You can explain what you offer when someone makes an enquiry. But that assumes they get that far. Your marketing reports won’t show how often an AI assistant leaves you out.

founded AI Wide Open, where she explores the practical impact of AI on businesses. In this guest post, she examines what an AI assistant needs to know about your business before it can recommend you.

You might still be deciding whether AI has a place in your business. Your next customer could already be using it to decide whether you’re worth considering. 👇🏻

The sale you never knew you lost

A customer could be ready to buy exactly what you sell and never find your business. They ask an AI assistant to recommend a supplier, choose from its shortlist and buy from a competitor without ever seeing what you offer.

You’ve spent years investing in search rankings and making it easier for customers to buy once they arrive. AI assistants are introducing another decision point before that visit. They can compare products or services, recommend what looks like the best fit and, in some cases, complete the purchase.

If your business doesn’t make that initial selection, you lose the chance to explain why you’re worth considering. There’s no abandoned basket to analyse or enquiry to follow up.

Your business simply disappeared earlier in the journey.

There’s no single AI shortlist

In traditional search, customers see a list of links and decide which ones to investigate. AI assistants make an initial selection for them, and that selection can vary considerably.

In September, John Lewis said searches originating from AI agents rose from 0.3% to 2.5% in a year, with the trend accelerating.

Euromonitor tested 1,000 shopping prompts across ChatGPT and Gemini in the UK and US. In the scalp-care category, the brand Nizoral appeared in 58% of Gemini responses but only 24% on ChatGPT. A brand that features prominently in one assistant’s recommendations can be far less visible in another.

Amazon’s Alexa for Shopping, which combines the former Rufus assistant with Alexa+, shows how much further this can go. Amazon reported that more than 350 million customers used the assistant in the preceding 12 months. It can answer product questions, compare prices and even buy something automatically at a price the customer sets.

That’s how AI is becoming a consideration layer, influencing which products customers evaluate before opening an individual listing.

When AI can't make sense of you

I think we’re getting too caught up in optimising for AI when businesses haven’t always got their basic product information right.

Read more

I know the feeling because I spend more time there than I’d like to admit. I publish something, keep up with people, get a reach number back and still have to work out whether the time was worth it.

That was where I started the latest Millennial Masters Live with , and . We kept coming back to what happens after somebody notices you, because that’s where the visible numbers stop helping.

LinkedIn looks most useful as the place where someone first encounters your work and decides whether you’re worth more of their time. 👇🏻

The follower count doesn’t tell you enough

Melanie started with the number LinkedIn makes easiest to see.

“The value of a follower is small unless they are taking another step.”

A growing audience can help more people become familiar with you, especially when your profile gives them a clear route to continue the relationship. The count itself doesn’t tell you whether they did.

Cory sees a narrower use for the number. His work includes helping people with content, so a visible audience and a consistent body of posts help support his credibility when a potential client checks him out.

He has also spoken with people who had more than 400,000 followers and were still struggling in business. In those cases, the content people saw had little connection with what the person was trying to sell.

That makes the follower number useful when it backs up the work you claim to understand. It can’t tell you whether the person who saw your profile ever became a client.

A reach number has the same limitation. LinkedIn can tell you that a post travelled without telling you what the people who saw it did afterwards. If you stop at the number, you’re measuring the part of the process the platform happens to show you.

New here? Subscribe to Millennial Masters for weekly reporting and conversations on the decisions behind building a business 📧

Millennial Masters

Helping founders make better business decisions

By Daniel Ionescu

Your profile is part of the sales process

Taylin uses LinkedIn mainly for cold outreach, so his profile has a specific job.

“My LinkedIn is more of a DM platform.”

He expects anyone considering his proposition to look at his profile before replying. His job is to make sure they find enough evidence to take him seriously.

Taylin runs a newsletter agency, so he puts his LinkedIn newsletter where a prospect can find it quickly. The writing gives them a sample of the work he sells. He also said he should publish more client proof and explain more of the way his agency works.

When he sends somebody a message, he assumes the profile check is coming.

“The first thing they’re gonna do if they’re considering working with me and taking my proposition seriously is look at my profile.”

Taylin judges the post by what happens when the person he contacted arrives there and decides whether he is worth replying to. Wide distribution matters less in that setup.

Melanie checks profiles carefully too. She looks at the headline and whether the person has taken the time to fill in their work history. She previously wrote for Millennial Masters about LinkedIn becoming part of your AI footprint because it gives people another place to check who you are.

A profile can do useful work before a sales conversation has started. It can answer some of the questions somebody has after your name appears in their inbox.

Attention needs somewhere to go

Melanie has built much of her business on conversations in LinkedIn DMs.

She uses the notifications people tend to ignore as reasons to get in touch. If somebody changes role or has a birthday, she sends a personalised message instead of pressing LinkedIn’s automatic congratulations button.

“It’s a complete conversation starter.”

A message is enough to count as progress when it begins a relationship that develops over time. Melanie also uses her LinkedIn newsletter to point readers towards longer writing on her Substack. She wants people to move from a social platform into a place where she can keep the relationship going.

Cory spent years focused mainly on LinkedIn. A couple of months into running a newsletter, he was already kicking himself for waiting.

“I’m already kicking myself saying, ‘Why didn’t I do this before?’”

A newsletter gave him direct access to the people who subscribed and control over what landed in their inbox. He could keep reaching them after the original LinkedIn post disappeared down the feed.

A post can introduce you to somebody. The next part depends on whether they have a reason to keep going.

Long-form gives people more time with you

For Taylin, the reason long-form matters is time under attention.

“You need them to consume as much of your content as possible in the shortest amount of time.”

He estimates that somebody who follows you on LinkedIn may spend only an hour or two with your posts across a whole year. Followers miss some posts and skim others.

An article or podcast can concentrate much more time into one sitting, with the podcast also letting people hear how you think without the compression of a short post.

Taylin saw the commercial effect in his own business. He said his sales increased significantly after he added a podcast, even though newsletters remain the centre of his work.

Melanie sometimes hears from people who remember an article she published months earlier even when they never liked or commented on it. They remembered enough to bring it up later.

Visible engagement can miss the reader who remembers an article months later and comes back when they need the work you sell.

Long-form also gives you room to prove more than familiarity. Taylin’s LinkedIn newsletter works as evidence for a prospect because writing newsletters is the service he sells. Melanie can show how she thinks through the articles she publishes. The content gives somebody more material to judge before they decide whether to spend money.

What counts as a result

Melanie’s test for personal content was whether the detail helps the reader understand something. Taylin’s strongest lead-generation example did exactly that.

His business went from making around 70k a month to a point where he couldn’t pay himself within 45 days after one mistake while he was away. He later wrote about what went wrong and what he changed after coming back.

He called it his best lead-generation post. Some of the attention became leads, which gave him a result he could trace back to the post.

When I look at my own time on LinkedIn, that gives me a more useful test. I still want a post to reach the right people. Once it does, I need to see whether anyone gives me more of their attention and whether that relationship produces something useful for the business.

If the number goes up and nothing happens afterwards, I still haven’t answered whether the time was worth it.

The idea came from going back through podcast interviews and pulling together quotes about founders selling their businesses. I started finding connections between their stories and realised I had enough material to bring them together in a film.

By the time I made it, I’d interviewed almost 100 founders and met some amazing people along the way. Thirteen of them tell their stories in their own words in my first documentary, The Exit Files: The Long Way Out.

I think what drew me towards video was what I noticed during the interviews. There’s something about seeing someone’s expression or hearing their voice change when they remember a difficult moment. You can tell when it still affects them, even years later.

I’m pleased with how it’s turned out, and I’m glad I chose video. Some of these stories really are better heard directly from the people who lived through them.

The documentary also became the starting point for my four-part Exit Files written series, which grew out of the research and editing process.

You can watch the full documentary above. If you prefer the big screen, it’ll premiere on my YouTube channel this evening. Both are in glorious 4K.

I hope you enjoy the documentary, and I’d love to know what you think. You can just leave a comment below.

While editing, a few things the founders said stayed with me. Some made me reconsider what a successful exit actually means, even for someone who’s made millions.

I picked out ten of those thoughts below. I think they’re worth reading if you’ve ever considered selling your business, or wondered what life might look like afterwards. 👇🏻

After the business is gone, who are you without it?

When Luke Tobin talked about going through a mourning process after selling his business, I recognised the feeling. I’ve lost a business myself, although my circumstances were very different. I remember having to work out who I was without something that had been such a big part of my life for so long.

I think part of what made me curious about these founders’ experiences was wanting to understand what it feels like at the other end of the spectrum. What happens when you sell for millions and get to leave on your own terms? Does having that financial security make it any easier to let go?

Hearing their stories, I was surprised by how familiar some of those feelings sounded. The financial outcomes couldn’t be more different, of course, and I wouldn’t pretend otherwise. Yet even after a hugely successful exit, you can still struggle with losing a business that helped define who you were.

I guess we’re all human, however the business ends.

You can leave that meeting excited about the work without asking whether it’s where your limited budget should go. That gets expensive when the question resurfaces weeks later and the team has to do the work again.

Martin Weigel spent four decades in strategy, including 14 years as Chief Strategy Officer at the advertising agency Wieden+Kennedy in Amsterdam. He’s the author of Strategy Is a Verb, and he knows how satisfying it is to win a client over.

I asked him to write about what you need to settle before giving an agency the go-ahead, especially when you love what they’ve put in front of you. 👇🏻

The room applauded, and the chief executive called it the best strategy presentation she’d seen in years. Nobody decided anything.

Sales in the brand’s established markets were weakening. The business saw room to grow elsewhere, where the brand was barely known, but it didn’t have enough money to invest properly in both. It needed to choose where to spend its budget.

The agency presented its answer in eighty slides, with insight and data alongside a call for the brand to return to its roots. A rousing manifesto film played in the middle of it. There was even a line, something about being the best of the best.

The agency built its answer around the existing markets without anyone choosing to put them first. Nobody asked what returning to the brand’s roots would mean in markets where it had none.

Weeks later, the client called the agency back to explain how the proposal would work in the new markets. That was the conversation they needed in the first meeting. The agency lost weeks repeating the work, and the client’s confidence took a knock. The agency built the deck to win agreement, and won it by leaving the hard question unasked.

I loved the big reveal

Joining Wieden+Kennedy Amsterdam as head of planning some 17 years ago opened my eyes to what a deck could look like. For the next 14 years, I sweated every slide and every pixel, beaming with pride at the gorgeous work my department delivered day in and day out.

I’ve given hundreds of presentations, and I confess that the theatre and drama of persuasion satisfied something in me. The big reveal is an adrenaline hit, but until we kick the habit, we’ll never put the deck down.

I know how easy it is for a presentation to take over the room. The speaker decides what appears next, and the audience follows along. An eighty-slide deck can give the impression that its authors have done an enormous amount of thinking. You can agree with the story without challenging the choice that matters.

Strategists are fond of saying that “strategy is sacrifice”. The hard part is deciding what matters enough to pursue and what we’re willing to leave out.

So why the industrial-scale slide count? Could it be that we feel compelled to make our case at painstakingly and gorgeously art-directed length because we worry (or know) that our audience just don’t trust us or believe us that much?

What that first meeting needed

Before you approve the work, ask what the proposal commits the business to.

Here, you’d ask: “Are we putting the money into our existing markets? What won’t we fund as a result?”

Ask the agency to explain why that’s a better use of the money than building the brand in new markets. If the recommendation depends on putting existing markets first, everyone needs to understand that before agreeing to it.

Once you’ve decided, write down what you’re authorising.

For example: “We’re using this budget to recover sales in our existing markets. Investment in new markets will have to wait.”

If you haven’t got enough information to choose, agree what you need to find out and who’ll bring it back. Make sure the team knows the spending decision remains open.

Thirty minutes in silence

I’ve experienced a different way. In 2004, Jeff Bezos banned PowerPoint from Amazon’s senior-team meetings and replaced presentations with written memos, initially four pages and later six. Participants spent thirty minutes reading in silence before discussing the document.

What felt like some kind of psychopathic torture test at the time makes more sense to me now. Writing a decent memo forces you to explain how your argument holds together. Everyone can read it at their own pace and go back to a claim that doesn’t add up, without a presenter pushing on to the next slide.

For your next proposal, ask the writer to put the decision you need to make at the top of the document. In this case, that was where to spend the available budget. The explanation underneath should compare the options and say why the writer recommends one. Give people time to read it before you ask them to agree.

Choose the meeting before the presentation

There are times when the job is to get people behind a direction you’ve already chosen. Theatre can absolutely play a legitimate and vital role there. The trouble comes when we use the same show to ask people to work out what they should do.

To my mind, there are three different kinds of meeting, and each calls for different preparation:

1. Hear what people know: bring your ears and give people room to speak.

2. Choose what to do: bring the options and explain how you’ll compare them. Let people challenge the recommendation.

3. Get people behind an agreed plan: use the presentation to help them understand and support the choice.

Only one of these might require the widescreen, Technicolour, all-singing, all-dancing “deck” that we cling to.

— A version of this article first appeared on Martin Weigel’s website. The client example is based on a real meeting, with details changed to protect the client.

This is part four of 🚪The Exit Files, my Millennial Masters series following founders through the years around a business sale.

Part one looked at the pressure of building towards a sale. Part two followed the work that continues after the deal. Part three explored how founders adjusted as the businesses they sold changed around them and after they finally left.

Several founders went back to building after selling. In this final instalment, I wanted to know what they chose next and whether the experience changed how they approached their next business. 👇🏻

The sales call she never wanted to make

Melissa Kwan found herself doing live demos for eWebinar, a product she built so customers could see it working and sign up without talking to anyone. The subscription cost around $100 a month. The calls took her back to the job she decided not to create again.

She sold Spacio, her real estate technology business, after nearly a decade of building companies in that industry. By then she knew what she wanted from the next one. She wanted a small remote team and a product she could sell online without spending her days at conferences or on calls.

eWebinar was meant to fit that life. When growth slowed, Melissa reached for something she knew she could do: sell. She spent months on calls with potential customers, even though the price of the product couldn’t support that kind of sales process. “I was getting on these calls to prove to myself that I could be productive,” she told me.

Melissa initially blamed the slowdown on her burnout. She later realised eWebinar’s marketing mainly reached businesses already convinced they needed automated webinars. She was reworking it to reach companies still satisfied with live webinars. Months of sales calls hadn’t solved that problem.

Trying to prove the next idea wrong

Nick Telson and his co-founder spent seven years building DesignMyNight before selling it for over £25 million. They stayed for another two years. When they began working on Trumpet, they wanted to keep building businesses without repeating the pace that took so much of their twenties.

They kept each idea under a project name while they took several possibilities to potential customers. They spent a month or two on each, looking for the reason it wouldn’t work. Nick would send a one-page outline and ask what people already used and why the proposed product might fail.

A fintech idea got good feedback. Nick and his co-founder still had to ask, “Are we the right people to run a fintech business?” Neither felt passionate enough about it to go ahead. Trumpet, a product for digital selling, passed both tests: customers saw a use for it, and the founders wanted the job that came with it.

They also invited Rory, a third co-founder, to be CEO. He could lead the team and be the face of the company while Nick and his original co-founder focused on the parts they wanted to own. Nick said they knew they didn’t have to work themselves to the bone this time.

Read more

During our Millennial Masters Live, Philip Hofmacher admitted he left one of his automations 80% finished for four weeks.

I recognised the problem immediately. I've got around eight projects in much the same state, often because I start something else before getting through the awkward last bit. I didn't think much about the unfinished backlog when I wrote about the new one-person company in June. 👇🏻

New here? Subscribe to Millennial Masters for weekly reporting and conversations on the decisions behind building a business 📧

Millennial Masters

Helping founders make better business decisions

By Daniel Ionescu

How much can one person run?

Behind Orel's setup are roughly 13 or 14 specialist bots, all reporting to one he calls Jarvis. He estimates that a feature he shipped in a week might previously have needed five developers working for a year.

“I just come up with an idea and I can test it an hour later and see how it works.”

Philip runs Write • Build • Scale with two co-founders and two employees, one of whom was part-time when we spoke. Their 11 or 12 bots handle specific jobs, including a daily check of the CRM for recent sales activity and conversations needing attention.

He doesn't think that amounts to a dozen virtual employees. The CRM check would take about an hour a day, nowhere near enough work to justify another hire.

“I prefer to have an AI do this one hour a day task for me rather than doing it myself or paying someone to do that.”

When building becomes too easy

After building two tech businesses, Cory Blumenfeld still caught himself making an old product mistake at BlueMoso. He began planning a portal around the functions it could have, then realised he hadn't properly established which problem customers needed it to solve.

“We're building before we actually truly understand the pain point that we're building for, right?”

With AI doing the coding, a new feature costs Orel a five-minute prompt and about an hour of checking, by his estimate. He also tested an AI coach with around 20 people, expanding the trial to 30 when others asked to join. Other experiments never reached general release.

“But testing it is so cheap that I don't see a reason why not.”

An agent working on WriteStack's landing page showed Orel the commercial upside. Visitors saw a button labelled “Start free”, then found they had to choose a plan and enter card details. Some abandoned the process. After he changed the button to “Start your free trial”, Orel reported a 10% rise in conversions. He still approves proposed changes before they go live.

Philip questioned where Orel's feature ideas were coming from. Some came from customers, while others were Orel's own.

The problem with 80% done

Philip picked up the 80/0 rule at a recent mastermind.

“If something is eighty percent done, it provides zero value.”

Philip left the final setup of his Threads reply bot unfinished before travelling. When we spoke, the bot still wasn't doing the job he built it for.

I have the same habit of leaving projects at the point where they need a final test or another round of feedback. Once something else takes my attention, I struggle to return to those open problems.

Cory recommended parking ideas on a list and revisiting them a few days later, though he no longer did it consistently himself. He paused work on his portal because he wasn't convinced it solved a clear customer problem.

Having more ideas became part of the problem. Philip thought “the most important skill nowadays is spotting the right opportunities rather than spotting ideas for growth.” Orel told Jarvis to file away new features and point him back to whatever he was supposed to be working on.

Someone still needs to know what's going on

Philip learnt his usual reply and close rates by doing sales work himself. He wants those figures before delegating the process to an agent, so he has something to compare the results against.

“I would not be able to judge if the AI is doing a good job or not.”

Even Orel said AI now writes so much of his code that he doesn't know 50 or 60% of it in detail. His development experience means he can still investigate a fault. Cory's unfinished portal raised a more worrying possibility: he didn't know enough to check whether the system stored customer data securely. Without that expertise, “you're just letting the AI pick for you”.

At Millennial Masters, I spent more than a month setting up the agents that handle sponsorship requests and editorial follow-ups. Much of that time went into writing the operating rules and correcting cases I didn't anticipate. The agents now draft replies that stay in my inbox until I review them.

All that setup makes the time saving difficult to calculate. I haven't measured how many hours my agents save after I count the corrections, and Philip couldn't give a figure for his business either. Some of his bots produce reports that nobody prepared before.

What still needs a human

Cory is building an agent to follow up overdue invoices and plans to have a virtual assistant oversee the system. That should free up the assistant to deal with conversations that need a person's attention.

“I refuse to hand over the customer relationships.”

The suggestion that an AI caller could chase people who missed his webinars made Philip recoil. He saw it as an intrusion into the trust he built with his community.

“I do not want an AI to call my people.”

His Substack welcome message is automated too. New readers are asked why they subscribed, and Philip or his VA takes over when they reply. He estimated his automated welcome message went out to 50 to 100 new readers a day. Orel automates welcomes as well, while refusing to turn comments into bot-written exchanges.

At that volume, Orel wondered whether a trained AI could do the same work as Philip's assistant: “what's the difference in this case between a VA and an AI that knows how you write?”

Philip tried AI-assisted replies before. In informal comparisons, he found his human VA's replies worked better for his readers. He wasn't ruling out a sufficiently capable AI, although training one to handle all those exchanges was more work than he wanted: “I don't have the time nor the expertise to train the AI to get it to a point where it really responds exactly the way that I would.”

Philip wished he'd introduced an AI check earlier to question whether a call or new project deserved his time before he committed.

I've already got an AI strategist helping me make decisions like that. The eight unfinished projects on my own list are still waiting for an answer.

She came to that view after moving from agency work and roles at Google and Sage into advising early-stage companies. She expected to focus on marketing, and the work soon pulled her back into the product because teams were sometimes trying to sell around assumptions they hadn’t tested.

She now works on those questions through ELVT Advisory. In one reporting-software project, buyer conversations showed that the team built something more complicated than the problem people thought they had. Those conversations led them to strip the product back.

In this episode, we get into how your own instinct can narrow your view of the market and how much weight to give enthusiastic early customers. We also look at the research that can expose a weak idea of who your customer is before more money goes into product or sales, including what to ask people who churned.

🔗 Find Ayse on LinkedIn, ELVT Advisory & Founder’s Stack Trace

Key takeaways

1️⃣ Early enthusiasm can mislead you

Early adopters are useful because they’ll tolerate an unfinished product and tell you what they want next. Their enthusiasm can also keep you building features the wider market doesn’t need. Treat those requests as evidence to investigate, then test them against the problem buyers are actually trying to solve.

2️⃣ Expertise needs outside evidence

Solving a problem yourself gives you a strong starting point, while also narrowing the range of solutions you can imagine. In the reporting-software example, buyers were already handling the problem with simpler workarounds. Use your experience to start the investigation, then find out whether other buyers see the problem the same way.

3️⃣ A job title won’t define your ICP

Two people with the same job title can be in completely different buying situations. Look at what has changed enough to create the need, then work out who else influences the purchase. A role name alone tells you very little about whether somebody is ready to buy.

4️⃣ Category creation has an education bill

Creating a category means paying to educate buyers before you can sell them the solution. That is a heavy burden for an early software business with limited resources. If the problem already exists inside an established market, focusing on a segment of that market can be a much cheaper route.

5️⃣ Talk to the customers who left

Start with people who still use the product, then make time for those who left. The second group can show you where expectations broke. Interview five current customers and five who left, using open questions and enough silence to hear the answers you didn’t expect. At one startup, former customers were offered a €20 Amazon gift card for a 15-minute call.

Also mentioned in this episode:

• The Kano model: A framework for prioritising product features by how they affect customer satisfaction.

• Steve Blank’s Customer Development Manifesto: The customer-development principles behind getting out of the building and testing assumptions with real buyers.

• Crossing the Chasm by Geoffrey Moore: The book behind the episode’s discussion of moving from early adopters to mainstream buyers.

In this episode

00:00 Episode highlights

01:51 Marketing wasn't the starting point

03:11 When buyers don't understand the product

05:26 Which features deserve to survive

08:24 The domain expert trap

11:59 From early adopters to enterprise

15:27 The market was bigger than expected

18:20 Diagnose the business problem first

24:02 Category creation gets expensive fast

26:43 Reaching buyers without blasting them

30:27 The 10-customer reality check

33:44 Where instinct still matters

Related episodes

This is part three of 🚪The Exit Files, my Millennial Masters series following founders through the years around a business sale.

In part one I looked at the pressure of building towards a sale. Part two followed the work that can continue after the deal.

This time, I’m looking at how founders adjust as the businesses they sold change around them, and after they finally leave. 👇🏻

Fourteen-hour days in Portugal

Luke Tobin moved to Portugal after selling Digital Ethos in a deal reported at £8 million. He had more money and more choice over how he spent his time. He still worked fourteen- or fifteen-hour days. “I should be out enjoying the Portuguese sunshine,” he said.

His partner asked why he still worked as though he were broke. Luke left home young and spent years fending for himself. He still feared he could lose everything.

“I still don't feel safe or completely comfortable.”

The sale changed Luke's finances much faster than it changed how secure he felt. He enjoyed his new life, but the fear that kept him working was still there.

He also enjoyed helping other people build businesses. Some of those long hours went into work he wanted to do. He struggled with how hard he kept pushing himself even after the financial pressure eased.

Watching the name disappear

The buyer brought Digital Ethos into the wider group and rebranded it within four months. Luke enjoyed working with the new team and learned from the integration. Watching the name disappear still hurt.

Digital Ethos was the most successful business he built. Winning awards brought recognition and a sense of achievement. Losing the brand meant losing a name he associated with that success.

He went through what he called a “mourning process” as he tried to find a sense of purpose after the brand disappeared.

You can agree to a sale and understand why the buyer wants to change the business. Living through those changes can still hurt. Luke knew the rebrand was coming, but it happened faster than he expected.

He knew he'd build again, which gave him something to look towards. He was grateful he managed to sell and still felt the loss of the business he created.

Read more

He left Scout Space, where he’d been CEO, to build Atlas Cup, a racing league for spacecraft. His plan puts rival teams into orbit and makes them prove what their technology can do in public.

Philip wants Atlas Cup to become an early performance test for the industry, even when a poor result could hurt businesses his own investors have backed.

I ask him how you sell sponsors on a sport nobody has watched yet, what founders should count as traction when revenue is years away, and why walking away from the CEO title proved harder than going without a salary.

🔗 Find Philip on LinkedIn

Key takeaways

1️⃣ Funding can hide weak proof

Philip points to space companies raising hundreds of millions before their technology has flown, with valuations climbing while the underlying technical evidence stays thin. When revenue sits years away, another funding round can start looking like progress. Founders and investors still need something that tests whether the product works.

2️⃣ Create an earlier test

Atlas Cup gives spacecraft companies a way to compete against the same conditions instead of reporting against their own private benchmarks. That matters well beyond space. If your normal traction metrics arrive late, find an earlier test that forces the product to perform somewhere customers, investors or partners can judge it.

3️⃣ Interest matters when people act on it

Atlas Cup had teams signing up, media approaching Philip and conversations running with four major global brands before the first race. Those signals matter because someone else is spending time, reputation or resources on the idea. Early traction gets more useful when it changes another organisation’s behaviour.

4️⃣ Don’t let fundraising rewrite the build order

Some investors encouraged Philip to raise $30 million and push towards human racing much sooner. He’s sticking with uncrewed racing first so the business can prove the format before taking on far more technical risk. More ambitious fundraising can open doors while also pulling the company towards a version it isn’t ready to build.

5️⃣ Status can be harder to give up than salary

Philip has gone close to a year without paying himself, yet he says leaving the CEO role was the bigger sacrifice. He’d reached the job people are supposed to want, then chose a business with a high chance of failure. Founder risk isn’t only financial. Walking away from an identity you’ve worked years to earn can be the harder decision.

In this episode

00:00 Proving which space startups are real

02:48 Leaving the CEO job to start again

06:04 Why space racing can work now

11:11 How do you race a spacecraft?

15:29 When funding starts to look like traction

24:08 Selling sponsors before the first race

29:51 Why he won’t raise $30m yet

36:41 The sacrifice wasn’t the salary

This is part two of 🚪The Exit Files, my three-part Millennial Masters series following ten founders through the years around a business sale.

In part one I explored the pressure of building towards a payout nobody could guarantee.

This time, I’m looking at the months or years of work that can follow a sale, and how the deal shapes the handover. 👇🏻

An hour after the money landed

The money arrived in Luke Tobin’s account on a Monday afternoon. About an hour later, he had a big presentation with the team. They didn’t know he’d sold Digital Ethos, and he couldn’t tell them yet.

The sale took about fourteen or fifteen months. Luke spent so long thinking about the figure that seeing it land in his account felt strange and anticlimactic. He still had a meeting to get through.

You can complete a sale and spend the next two years working in the business. The deal determines how much money you still have to earn and when you can leave.

The Millennial Masters I spoke to agreed to very different terms. Their experiences show why it’s worth looking closely at the job you’ll have after the sale, including who gets to make decisions about the business you built.

Another two years on the job

Thibault Louis-Lucas, known as Tibo, and his co-founder sold Tweet Hunter and Taplio in a deal that could reach $10 million. He expected to start building something new almost immediately, but buyers wanted him to commit another two years.

“Something that we didn’t anticipate is that all the buyers out there were asking us to stay two years in business with the acquisition.”

Staying meant spending two more years growing businesses he no longer owned, with much of his payment depending on how well he did.

The buyer paid $2 million upfront, with up to $8 million more tied to revenue milestones over the following two years. A headline price can include money you’ll only receive if the business keeps growing.

Tibo knew the buyer’s founder from school. The team was relatively small, and he felt they understood how he liked to work. In our conversation, he credited the freedom to keep building independently with helping the arrangement run smoothly.

They hit five of the six revenue milestones, so the total sale amount came to $8 million. The teams stayed quite separate, though, and Tibo thought closer collaboration could have helped them grow further. The independence he wanted also limited how much they worked together.

You’ll be working with the buyer while pursuing those targets. How much control you keep over the work deserves as much attention as how long you agree to stay.

Read more

I can catch a bad edit or an automation misbehaving because I know how it should work and when the output looks wrong. But when I use AI in an area I don’t know that well, there’s no safety net.

I’m relying on an answer I’m not really qualified to challenge, especially when ChatGPT or Claude can sound so reassuringly confident and convincing.

spent years working in AI in Silicon Valley and on Wall Street before launching her podcast, Artificial Reality. Now she asks operators across industries about the AI they use and what they’d pay for that nobody has built yet.

Because nobody on her show sells AI, they answer honestly. And in those conversations, she spotted the cost nobody puts on the AI invoice: you still need enough expertise to catch a bad answer. 👇🏻

A husband-and-wife team walked into an M&A meeting holding a $5 million valuation of their business. It had made $125,000 in net profit that year, with no intellectual property or real estate. Claude had produced the number, and they believed it because nothing in the report suggested they shouldn’t.

Christine McDannell, who runs the sell-side M&A firm The Magnolia Firm, knew it was wrong before she finished reading. She told them the multiple for that business was three to four times net profit, emailing them before the call because some things shouldn’t be said to someone’s face on Zoom.

Then she read the report behind the number. Claude had assigned a dollar value to every department in the company based on payroll. It valued the accounting department at $200,000. It had counted expenses as assets. That’s how it arrived at $5 million.

The real problem is who can catch it

Christine caught it in seconds because she values businesses for a living. The sellers couldn’t, because they don’t. They’d used a capable tool, presumably prompted it well, and received a confident, well-formatted document about the most consequential financial decision of their lives.

They’d bought precisely the thing they had no way to check. It points to a cost of this technology almost nobody prices in:

The expertise you need to check an AI’s work is roughly the expertise you were hoping it would replace.

Follow that where it goes. The safest uses are the Tuesday admin you could’ve done yourself, because you’d spot an error instantly.

The risk rises when AI moves into a once-a-decade decision in a field you don’t practise, where you may never know it got something wrong.

Millennial Masters

Helping founders make better business decisions

By Daniel Ionescu

The failures rhyme

A charter client told his captain it was safe to anchor in 25-knot winds. He was certain because ChatGPT had said so. He’d screenshotted the answer and circled it in pen before the conversation started.

Jennifer Kerum, who runs a luxury yacht charter in Croatia, was the one who had to repair that relationship afterwards, not the model. Her line: “I don’t care what ChatGPT says. You need to trust the captain.”

Why did the model say yes? Because whether it’s safe to anchor depends on the seabed, the shelter of that particular bay, the boat, the crew and the forecast at four in the morning. That knowledge lives in captains. It isn’t written down anywhere on the internet.

Nikki Siso opened a restaurant in Austin. She ran a 45-page lease through ChatGPT, got eight risk flags in three minutes, and had her lawyer review it in 10 minutes for free instead of billing her $1,000. Steal that one, it works.

But nothing told her the sequence: that the mechanical, electrical and plumbing engineer has to come before the permit application. That cost her a month and $30,000. The sequence isn’t written down either. It lives in the head of the inspector, the contractor and whoever opened a restaurant in that town last year.

Sofiane Ghorbel, with 25 years in luxury hospitality, gave me the sentence I’ve quoted in every talk since: “We don’t sell rooms, we sell trust.” I asked him about Marriott’s RENAI virtual concierge, which struggled with a guest’s request for a nearby Italian restaurant with bar seating, so the guest was sent back to the human concierge. Sofiane said AI was already proving more useful behind the scenes in forecasting, revenue management and operations.

Line the cases up and the shape is hard to miss:

Across the seven industries I looked at, every AI win was invisible to the customer. Every failure was visible.

The knowledge AI doesn’t have

I spent two and a half years deep in Silicon Valley. The engineers I worked with were exceptional, but they’d never done the jobs they were building for. They hadn’t walked a factory floor or sat with someone being told what their life’s work is worth.

So the tools ship generic answers. They can cover a huge amount of ground without having the experience that comes from doing the job.

Those legacy industry firms doing twenty million a year on sticky notes have looked at the offer and decided it isn’t good enough.

She grew up working in her family business, but the wake-up call came when someone with a finance background diagnosed what was wrong in just two minutes. That was when Lauren knew finance was the language she needed to understand what was really happening.

Knowing the numbers didn’t make tough business decisions painless. For Lauren, they made the reality harder to avoid when she had to close her own startup, Roost. That path took her into CFO work with startups, and today she advises founders through Lauren Pearl Consulting. She also co-hosts The Growth Minded CFO podcast.

In this episode, we get into how to build a financial model that tests whether the business actually works, and what the numbers can expose before a big decision. Lauren also explains why hiring costs more than you expect and where growth forecasts start to go wrong.

🔗 Find Lauren on LinkedIn

Key takeaways

1️⃣ Understand before you delegate

Hiring an accountant or finance lead doesn’t remove your responsibility for the numbers. You don’t need to do every calculation yourself, but you should understand enough to question the answers and know when something doesn’t add up.

2️⃣ Growth needs work behind it

A forecast shouldn’t assume revenue rises because you want it to. Lauren ties growth back to things the business can actually do, then updates the assumptions as real results come in.

3️⃣ Hiring costs more than salary

A £100k hire can become a much bigger commitment once you include equipment, tax, HR costs and the cash delay before that person starts generating revenue. The first hire can be especially expensive.

4️⃣ The first model tests the business

Before the business exists, Lauren sees the financial model as a feasibility study. Its job is to show whether realistic revenue can cover the costs and whether the business has enough room to work.

5️⃣ Numbers can shorten the denial

Lauren knew Roost needed to close before she actually closed it. The numbers won’t make that kind of decision painless, but they can make the warning signs much harder to ignore.

In this episode

00:00 Introduction to Lauren Pearl

01:07 Where finance became real

09:50 What founders still need to understand

20:01 Growth without work

35:04 Deciding what failure looks like

41:42 Why the conversations matter more

43:57 Where AI gets finance wrong

50:48 Start with the question

53:37 The hidden cost of hiring

57:32 Can this business make money?

62:47 Learning finance is never wasted

I’ve spoken to enough Millennial Masters who’ve sold their businesses to know how much getting to an exit takes out of you.

Ask them what the years around the sale felt like and you’ll hear about exhaustion, or going back to work with millions already in the bank.

🚪The Exit Files series follows the experiences of ten founders, from hypergrowth, through to the sale and into life after.

First, the years spent building towards a payout nobody could guarantee, and what they really cost. 👇🏻

The sale is still years away

Luke Tobin was winning work faster than his agency could look after it. Digital Ethos was doubling in size each year, heading towards almost 100 people in five years. He was trying to hire enough people and put systems in place to deliver the work he’d already sold.

“Honestly, it was always a bit unstable.”

Every new client creates work that somebody has to deliver, and hiring someone to do it takes time. When the business doubles again, whatever worked last year is already stretched.

He eventually sold the business for a reported £8 million. Knowing that makes the pressure easier to justify in hindsight. He had to keep running the business without knowing whether the sale would go through. The strain can start years before you negotiate a sale.

Keeping up is hard work

Matt Watson lived it at VinSolutions, which made software for car dealerships. During the financial crisis, dealers were looking for ways to cut costs and sell cars online. That created demand for his business at a time when others were struggling.

“We were growing way faster than we could even keep up.”

VinSolutions reached $30 million in annual revenue without outside investment. But it didn’t have enough servers or support people. Matt could see how a cash injection would help the team hire ahead of demand. Eventually, profits grew enough that they no longer needed outside investment.

Revenue makes a company look further along than it may feel when you’re running it. A customer can sign today and need help tomorrow. Recruiting and training someone to provide that help takes longer. Until then, you and your existing team take the hit.

Luke’s agency and Matt’s software company were different businesses, yet both had customers arriving faster than they could serve them. Keeping up was exhausting, and every new sale added more work to an already stretched team.

Getting bigger doesn’t guarantee this will ease. More demand puts more pressure on your team while customers still expect everything on time.

Read more

Hundreds of emails later… and the replies don’t come. So you send more because there are always more names waiting.

The list keeps you busy enough to avoid asking why any of those businesses would want to hear from you today.

That’s why I was intrigued when , who’s obsessed with finding useful things to do with Claude, explained how he finds the names on that list with an actual reason to buy now. 👇🏻

In July, a client rang me. He sells scheduling software and pays for one of those B2B data tools that does what the demo says it will.

He got the business to seven figures without a sales team. He shared his screen. There were 500 companies sitting in a spreadsheet, every one matching his filters.

I asked him who he was emailing first. He went quiet for a second. “I’m working through them alphabetically.”

His tool did its job and handed him the names. He had no way to pick between them, so he sorted by column A and started typing.

A match is not a reason

A match tells you a company looks like your customer. To prioritise it, you need to know what changed there recently.

Software finds matches brilliantly. Spotting useful signals takes more work.

Your data tool knows a company has 80 staff, sells software and is based in Manchester. It has no idea the operations lead walked out six weeks ago and nobody replaced her.

The operations lead leaving gives you something useful to open an email with.

The same filters produce the same list

You pick an industry, headcount band, region and maybe a tech tag. So does your competitor.

You run the query on Tuesday. He runs it on Wednesday. You both get the same 500 companies, and then you both email them.

Your advantage is the reason behind the list.

That reason is in data no vendor holds: your closed deals, call recordings, the customers who bought inside a fortnight and the ones who took nine months.

Your competitor cannot see any of it.

What your sales tool can’t decide

Build the list from your own history

Put your last ten to twenty closed deals in one place: call transcripts, CRM exports, email threads and whatever notes you kept.

I used Claude Code to read a local folder containing eighteen deals in one go.

Then ask one narrow question: For each of these companies, what changed in the 90 days before they first replied to us?

Use the deal material to find the first clues, then check them against public sources. Tell it to look for hires, launches, new sites, leadership moves and tooling switches.

Ask which changes appear most often. You’re looking for what was happening inside those companies before they bought. Build the pool using the data tool you already pay for.

Use broad filters at this stage. You want a few hundred companies because this is only the starting pool. Take the pool and score every company against the changes you found in your closed deals. Ask for one line explaining each score. Only award a score when you can show the evidence behind it.

The evidence should come from places you can check, such as job boards for hires, local press and planning notices for new sites, or the company’s careers page for the volume of open roles.

That one line does the heavy lifting. It becomes the opening of your email, and it answers the question he couldn’t answer in July.

Weak buying signals

Some changes create more noise than buying intent.

Funding rounds attract sellers quickly, so the buyer’s inbox fills up at the same time. A raise shows that money has arrived. It doesn’t reveal a problem your product can solve.

Headcount growth is too broad to tell you much. Look for the hire that points to a specific gap.

Check the age of the evidence too. A role posted in March is filled or forgotten by July. Signals go stale, so set a window of about six weeks and bin the rest.

What happened when we rebuilt the list

For the scheduling software company, we rebuilt the list around three signals tied directly to the problem it fixes.

• A new Operations Manager or Regional Director, because new leaders pull apart old systems in their first 90 days.

• A second site announced or leased, which is the point where spreadsheet shift planning stops working.

• Five or more part-time or hourly roles posted at once, which means the workforce is growing faster than anyone can schedule it.

About 4,000 companies fit the broad filters. In any month, a few hundred showed one of those changes. Those few hundred became the list, and for the first time we had something useful to write about.

Each email opened on the change found at that company. The volume barely changed: roughly 1,000 emails in June and another 1,000 in August.

August is the worst outbound month of the year in Europe, which makes the increase harder to dismiss. His reply rate went from 0.9% to 7.6%. We filtered out bounces and out-of-office responses, so every reply was written by a person.

Cleaner addresses will have helped the reply rate. The copy got better because the list gave him something to say.

Find your reason to write

Pick one company from the list your data tool gave you this morning and say out loud why it needs you this month.

If you can’t answer, open your last ten won deals and look for what changed before those customers bought. Find twenty companies where that same change is happening now.

👤 Jasper Vanuytrecht is a fractional GTM lead who helps B2B founding teams use Claude Code to research prospects and build outbound systems they can run themselves. He has spent a decade working in sales and writes Founder’s GTM, where he shares practical ways to win new clients without hiring more sellers. Connect with Jasper on LinkedIn.

Founder's GTM

The Claude GTM skills founders use to win new clients without hiring more sellers. [+ Claude Code Starter Guide upon sign-up] ⬇

By Jasper Vanu | Claude for GTM

That split us pretty quickly in the latest Live I hosted with , , and .

Alicia thought obvious AI copy from someone she knows looks lazy. Anjeanette’s first instinct was more generous: maybe they were just busy. Cory admitted some of his earliest writing was “fully AI slop”, but said it gave him the confidence to start.

The argument really came down to standards 👇🏻

When does saving time with AI become an excuse for work you wouldn’t have accepted from yourself before?

When anyone could have written it

Anjeanette calls it the “robot voice problem”.

“If literally any other human in my position could write this post, then it’s probably not good enough.”

For Alicia, the problem is copy that says plenty without saying anything specific:

“AI says something that anyone could have said. Oftentimes it says a lot without saying anything whatsoever.”

She asks whether you’d actually talk like that to someone over lunch.

The copy can read perfectly well while leaving almost nothing recognisable about the person publishing it. Something recognisably yours should survive the process.

And you still need enough judgement to notice when the person has disappeared from the work.

Can AI teach you to write?

Alicia thinks people who are still learning to write should be particularly careful.

“If you don’t know what AI copy looks like, you don’t see it. Then it’s really hard for you to take the AI back out of it.”

Cory disagreed based on his own experience. He used to think he was a terrible writer, and AI gave him somewhere to start.

“The content I used to put out was just fully AI slop. But it gave me the confidence to get going.”

He used it to understand how other writers structured things and learn techniques he hadn’t picked up before. He doesn’t rely on it in the same way now.

Claudia is comfortable letting AI into the process earlier.

“It really comes down to the human editing.”

She’s happy for AI to help with an early draft, provided she goes back through it, changes the tone and removes what doesn’t belong.

Cory learnt as he used it. Alicia’s concern is what happens when somebody never builds enough judgement to know what needs fixing.

Starting with your own material makes it easier to keep yourself in the draft.

Get the next edition on AI and better business decisions 📩

Millennial Masters

Helping founders make better business decisions.

By Daniel Ionescu

Start with something that’s already yours

Cory now starts with material that already belongs to him. He records conversations, pulls transcripts into his system and uses AI to find ideas in things he has actually said.

“I don’t want my content to be written by AI. I want it to be the actual words that I say.”

Claudia often starts by dictating an idea into AI and using the conversation to think it through rather than asking it to produce something from nothing. I prefer that too because the model gets your context and natural phrasing before it starts drafting.

I’ve ended up doing much the same with my inbox. I recently connected ChatGPT after spending an unreasonable amount of time building SOPs and rules for how I deal with guest pitches, requests and follow-ups.

The first draft follows the rules I already use. I still review it, and if the action is wrong or the reply doesn’t sound like me, it doesn’t go out.

That source material matters. A transcript or voice note gives AI words and thoughts that already came from you. An SOP gives it decisions you’ve already made instead of asking it to guess how you behave.

Seven writers started sounding like one robot

At one point, Anjeanette had seven writers working across completely different industries, and their work started sounding strangely similar.

“It didn’t matter what industry it was. All of a sudden, the same stuff started coming in.”

She was fine with them using AI.

“I don’t care if you use AI, but don’t let me know that you used AI. Trick me.”

What frustrated her was that some writers had discovered they could produce in ten minutes what previously took two hours, then hand over the first thing AI gave them.

“If they had just spent the time to master the tool and get a little better at it, I would never have known.”

Their output pushed her to learn the tools herself. Eventually she started asking why she was paying people for work she could produce better on her own.

What cost them was handing over work that made the shortcut obvious.

Sometimes the right move is not to post

I asked when they’d rather skip a post than put their name to something AI created.

Anjeanette’s line is whether she can still recognise the person in the work:

“If you can’t recognise yourself in it, or your client in it, I’ll scrap it.”

Alicia had posted on LinkedIn that day for the first time in a week because she hadn’t had the headspace.

“I would never put my name to something just for the sake of posting content.”

Claudia’s test is whether something personal still made it into the work.

“I draw the line when there’s no personal touchpoint.”

That could be something she’s experienced or actually happening in the business. AI can help shape it. She still wants something there that belongs to the person behind it.

Humans don’t have a protected monopoly on creativity that AI will never touch.

A lot of what we call original work is already a remix of things we’ve seen and learnt from before.

AI can absolutely throw out an idea I hadn’t considered, even if nine of the ten suggestions are mediocre at best.

Fine. I’ll use the spark. Once the draft exists, I still want to decide whether it’s something I actually believe and whether the standard is good enough for my name.

If my name is on it, I want the judgement behind it to still be mine.

📡 Live Discussions

has seen that problem from the inside. She built and sold an agency, later ran the company that acquired it to £14m revenue, and now writes The Humans in the Loop and runs The AI Edit.

In this guest post, she looks at what happens after AI gets switched on, when the model changes, the business moves on or the system simply stops without anybody noticing. 👇🏻

If you hired someone and never looked at their work, you’d call that negligence.

It’s surprisingly easy to do the same thing with AI. Something gets set up, tested and switched on. It works, so everyone moves on.

Three months later, the model may be different and the process around it may have moved on. The information it relies on may no longer be current, or the workflow might have stopped without producing the kind of error that gets anybody’s attention.

You only find out when somebody eventually notices. We’re putting more AI systems into businesses without always deciding who owns them once they’re live.

New here? Millennial Masters shares practical lessons from founders on building better companies and dealing with the parts of the job nobody prepares you for. Subscribe for the next one 👇🏻

Millennial Masters

Helping founders make better business decisions.

By Daniel Ionescu

The model can change underneath you

When you build an AI workflow, you test what comes out, decide it’s good enough and move on. But not every model name points to something fixed forever. OpenAI’s chat-latest model, for example, points to the latest Instant model used in ChatGPT. OpenAI says the underlying model snapshot is regularly updated.

We saw another version of this when OpenAI retired GPT-4o from ChatGPT in February 2026. GPTs using retired models were automatically moved to newer equivalents. OpenAI was clear that its API wasn’t changing at that point, so this wasn’t a case of every Make or Zapier automation being silently switched.

Something you tested and approved can still end up running on a different model or version later. The output may be perfectly fine. Unless somebody checks it again, you don’t know.

The technology doesn’t have to change for the same problem to appear. Sometimes your business changes instead.

Your process changed, your AI didn’t

The production process for my Insider’s AI Briefing ran through a series of automations inside Notion. They took me half a day to build and were worth every minute, until I changed the thing they were automating.

I moved the session from Zoom to Substack Live. The way I set it up changed, but the automations didn’t. They carried on asking me for Zoom links when what I needed was a Substack Live link. They were doing exactly what I’d told them to do. The instruction had simply expired, and nothing in the system could tell me that.

I know they need updating. Fixing them takes time I haven’t found yet, so I’m running on substandard automations and I’m the person whose job it is to notice. There’s a harder version of this problem when the workflow stops doing anything at all.

The workflow stopped and nobody knew

I run a Zap that watches a folder in Outlook and turns anything I drop in there into a task. Then I migrated my email, which reset the Zaps attached to it, and that one stopped.

I didn’t know. There was no error message because, from Zapier’s point of view, nothing had failed. It simply wasn’t being asked to do anything anymore.

Three weeks later, someone chased me about a guest post I owed him. I went looking and found the task sitting in the Outlook folder where I’d left it. There were nine others behind it.

This is one of the hardest failures to catch because there’s nothing obvious to investigate. A wrong answer will usually reveal itself eventually. Missing work can sit there until the person waiting for it asks where it is.

A system can also keep working while the thing it knows slowly goes out of date.

Nobody was keeping it current

A client had a custom GPT trained on the company’s tone of voice. It was good, and the team used it every day for client-facing work. Then the person who built it left.

The GPT still worked and sounded roughly right. The problem was that the company kept changing while the GPT stayed where it was.

Nobody had decided who was responsible for keeping it current because they’d treated it as something that could be set up once and left alone. It never stopped answering, so there was no obvious moment when somebody had to question it.

What sounded exactly right in January was merely adequate by August, but it still sounded just as confident. Keeping a system current still doesn’t guarantee the output is right.

Plausible isn’t the same as right

AI produces work that often looks finished before anybody has properly checked it. The language is fluent, the formatting looks clean and there may be nothing obvious telling you to stop. That makes it easy to trust too quickly.

Even Deloitte got caught by this. It partially refunded the Australian government after a report produced using an Azure OpenAI model contained fabricated citations and a made-up court quote. The problems weren’t caught before publication. A University of Sydney researcher spotted them afterwards.

If work like that can get through Deloitte’s checks, looking finished clearly isn’t enough. And when it goes wrong, the AI doesn’t get embarrassed. You do.

So who owns it?

For most small businesses, this doesn’t need to start with another hire. It needs a name against each system.

Somebody should know what it was built to do, check what it produces often enough to notice when that changes and have the authority to switch it off.

For every AI system running in your business, ask:

• Who owns this?

• When did they last look at what it produced?

• What would have to happen for them to notice it had stopped being good?

If nobody’s name is against it, it isn’t a system you’re running. It’s a system that’s still switched on.

👤 Heather Baker is the creator of The Humans in the Loop and founder of The AI Edit, helping leaders understand and use AI in their businesses. She previously built and sold an agency before becoming CEO of the company that acquired it and growing the business to £14m revenue.

The Humans in the Loop

Helping leaders think clearly about AI.

By Heather Baker

Sophia Matveeva, founder of Tech for Non-Techies, learned that after raising money for her first startup and hiring a technical team. She was the CEO, but when a developer started talking about building the back end on Ruby on Rails, she realised she had no idea what he meant.

Her first instinct was to learn to code. She took computer science and Python courses before seeing she was spending hours learning skills that wouldn’t help her do her actual job. She needed enough technical understanding to make decisions, ask useful questions and work effectively with the team.

AI has lowered the barrier again. A non-technical founder can now make a convincing prototype in an afternoon, which makes it easier to start before finding out whether people understand the product or would pay for it.

Sophia argues that this is where founders need more discipline. You can use AI to get an idea in front of people much earlier and learn from what they actually do with it. Once you’re making something people will pay for, you still need people who understand the technology well enough to know what’s happening underneath.

In our conversation, Sophia gets into what non-technical founders need to understand about tech, how to test an idea before spending heavily on it and where vibe coding stops being enough.

🔗 Find Sophia on LinkedIn

Got an idea you want to build? Subscribe for more on finding out if people actually want it first 👇🏻

Millennial Masters

Helping founders make better business decisions.

By Daniel Ionescu

Key takeaways

1️⃣ You need to speak tech before you lead it

You don’t need to become a developer, but you need enough understanding to work with one properly. That means knowing what’s being built, why something takes time and what you’re giving up when you choose one feature over another.

2️⃣ Test the idea before you spend money building it

Sophia recommends putting a prototype in front of potential customers before writing production code. Watching how people actually use it can expose problems that are much cheaper to fix at that stage.

3️⃣ Don’t ask customers to validate your idea for you

People often tell you something sounds good because they want to be helpful. Sophia recommends approaching user interviews with curiosity, giving as little guidance as possible and making people comfortable telling you what they don’t understand.

4️⃣ Vibe coding is great for testing, not the finished product

AI lets non-technical founders create convincing prototypes much faster and cheaper. Sophia warns against assuming that means you can safely build an entire commercial product yourself when you can’t judge the quality of the underlying code.

5️⃣ A better product can still lose

Customers already have tools they know and systems built around them. Getting them to switch can require much more than improving the product, particularly when an established competitor is deeply embedded in how a company works.

In this episode

00:00 Introduction to Sophia Matveeva

02:08 Is vibe coding improving product testing?

05:34 Why founders need to speak tech

12:54 How to manage developers

15:31 Test your product before building it

18:47 How to get honest user feedback

22:47 Will people actually pay?

25:57 How to challenge your technical team

30:38 Choosing which features to build

32:57 Why Sophia shut down Enty

35:22 Funding versus bootstrapping

37:23 When to stop funding your own business

39:19 Why better products still lose

42:58 What AI changes for non-technical founders

45:47 Building is easier, finding customers isn’t

The trouble arrives a few years later, when you’re hiring experienced people after giving fewer juniors the chance to learn the job.

Jordan Stachini sees the incentive clearly in a creative agency. If AI can generate ideas quickly, the obvious cost-cutting question is whether the agency still needs ten junior creatives.

The best people she’s worked with became good through trial and error, moving from junior roles towards becoming middleweight designers, lead designers and creative directors.

“How do you then get more senior creatives coming through if you’re stopping it right at the start?” she asks.

If you remove more of the work people used to learn on, you need another way to build the experience you’ll still be hiring for later. 👇🏻

Junior work is already changing

Nick Holzherr says companies are already changing who they employ. “They’re not hiring junior marketing people anymore,” he says. “They are hiring more senior people that can use AI. So they just need fewer people.”

The appeal is obvious: an experienced marketer who knows how to use AI needs less supervision and is better placed to judge what AI produces. You get useful work sooner without paying to teach several people from scratch.

The problem can also appear when the job survives. Konstantin Klyagin’s software agencies use AI to take boring work away from existing developers and leave them extra time for more interesting work. He doesn’t frame it as a way to replace staff.

Konstantin says building larger, reliable software still requires somebody who understands architecture, infrastructure and how the system fits together. If AI removes some of the repetition that helped people develop that knowledge, your company must decide how a junior developer is going to learn it.

The boring work was teaching you something

Nobody needs to spend years copying numbers between spreadsheets because boredom supposedly builds character. Some tasks deserve to disappear.

Early work also puts somebody in front of hundreds of small decisions. A junior writer starts noticing why one draft is weak. Somebody handling clients learns to recognise the moment confidence begins to slip. Those lessons add up before the person can fully explain what they’ve learnt.

Anjeanette Carter argues that AI is far more useful when somebody can combine it with knowledge of the subject. Without that, a junior can get a better first draft before they understand why it’s good or recognise when it’s poor. They then learn by reviewing the work and seeing how it performs for a real client or user.

Millennial Masters

Helping founders make better business decisions

By Daniel Ionescu

You used to learn by overhearing

Michael Buckworth saw what happened to trainee lawyers who worked remotely during the pandemic. Junior lawyers struggled, stopped developing and produced poorer work. He says less contact with senior lawyers put pandemic trainees at a disadvantage. He called it the loss of “osmosis”: hearing senior people speak to clients and listening to conversations around the team.

Nick Telson-Sillett noticed the same mechanism in sales: junior employees sitting nearby picked up his intonation and small techniques simply by hearing his calls. Recordings preserve the call, but not all of that everyday exposure.

With AI handling more of the practice work, a remote junior will spend less time around experienced colleagues. If your team is remote, shared documents and a weekly call won’t recreate all the learning that once happened around experienced people.

The senior hire is the easier choice

Gavin Bell’s first hire shows what close apprenticeship can produce. Keir arrived raw. They sat side by side every day handling clients, giving Gavin plenty of chances to explain his approach and correct the work as it happened. Keir developed through the experience.

Gavin now has the money to hire senior people who can work independently sooner. Teaching another inexperienced employee would cost time and attention that the business no longer has to spend in the same way.

There’s nothing hypocritical in that decision. The same choice becomes more attractive once you can afford experienced people. The problem grows when companies across the market make the same decision and give fewer juniors the years of work needed to become the senior hire everyone wants.

AI can help rebuild the practice

Ben Tasker found a useful role for AI inside training: using it for roleplay so somebody can build “muscle memory” and learn common cues before they face a live client.

AI can give juniors more chances to practise when the job itself gives them fewer. They can rehearse a difficult conversation, try another response and discuss what happened with an experienced colleague. Asking AI for a finished answer teaches less than requiring the person to choose an approach and explain it.

That’s why AI makes practice cheaper and easier to repeat. But experienced colleagues still define what good looks like.

Training becomes somebody’s job

Jordan already treats this as a cost the company has to absorb. She hires 19-year-old apprentices, pays them a professional wage and thinks of it as a bigger investment because the employer has an obligation to teach, mentor and support them.

That matters more as accidental training disappears. Universities can prepare people for work, but the judgement specific to your company has to grow inside the business. If you use AI and run a distributed team, juniors already get less of that everyday practice.

Look at the experience disappearing from a role as AI takes over more tasks. Remove the genuine waste. Before it goes, check whether any of that work was also giving people useful practice.

Once you know what still matters, put it back into the role on purpose. Give a junior real responsibility with AI assistance and keep them close to the consequences. Let them hear experienced colleagues handle clients before you expect them to lead those conversations themselves.

AI can help you cut entry-level jobs and reduce how much junior work you need today. Five years from now, you’ll still be looking for experienced people. The senior hire you want then has to learn somewhere first.

It often comes back as something somebody still has to review or correct. Models have improved, but in my experience, they still need checking.

That’s why , the entrepreneur behind Taking My Time Back, uses a simple three-question test before handing work to AI.

In this guest piece, he explains where human judgement still matters when AI is doing more of the work. 👇🏻

“The AI told me to do this.”

I hear that from clients and my own team more often than I’d like. My first response is usually: did you think about it? Did you actually read it?

The problem isn’t always the answer AI gave them. Sometimes the mistake is asking AI to make the judgement for them.

If you don’t know what good looks like, almost anything AI gives you can seem good enough. You need to understand the work well enough to tell it what you want and spot when it gets something wrong.

Millennial Masters

Helping founders make better business decisions.

By Daniel Ionescu

The three-question test

I use AI every day. For repetitive work, it makes sense because it gives me more time to think. But there’s a phrase I keep coming back to: cheap delegation gets expensive.

The cost usually shows up later, when somebody has to fix the output or a customer notices the difference.

So before I delegate a task, whether it’s to AI, a virtual assistant or somebody else, I think about three things.

What I hand over

AI isn’t going to do my podcasts, but it can handle plenty of the work around them.

Across the team, I’d estimate we get back an hour and a half to two hours a day from repetitive work that didn’t need much thought in the first place.

After one of my calls, I want the transcript saved in Notion, turned into notes, summarised and used to draft a follow-up email.

A person used to have to know the call had ended and work through those jobs one by one. Now the workflow runs automatically, without me having to kick off each step.

The problem is that a bad automation can keep producing poor work every time it runs, and somebody still has to catch it.

I’ve seen this in outsourcing too. When somebody spends every day doing the same mindless task, they’re not learning much from it. That’s the sort of work I’m happy to hand over. I’m more careful with work built on years of experience.

Ford had been relying more heavily on automated quality systems, but executives weren’t getting the results they wanted. It brought in 350 veteran engineers to help find problems and improve the AI itself.

The review bill

At what point does reviewing work that AI created cheaply and quickly take longer than doing it properly in the first place?

It’s easy to skip that question until the review work starts piling up.

AI can save a lot of time, but some of that saving gets swallowed by keeping things working. A model update can break an automation or suddenly make old instructions less reliable.

That rework adds up. In one survey, 85% of employees said AI saved them 1–7 hours a week, but 37% of the time saved went back into correcting, rewriting or adding missing context. Often, the clean-up falls to whoever notices the problem and has to sort it out.

Saving time doesn’t necessarily mean less work either.

I can take more calls because I’m not writing every summary myself, but those calls create more follow-up and give my team more clients to support.

AI lets the business take on more, but the workload doesn’t shrink on its own.

The time you save is easy to fill again, so you have to be deliberate if you want any of it to stay free.

Your name’s still on it

If something goes out under your name, it’s still your responsibility, whether AI touched it or not. You still have to verify the information and decide whether the work meets your standard.

Air Canada learned that the hard way when its chatbot gave a customer incorrect fare information. A tribunal held the airline responsible for what the bot had told him.

With writing, I often start from a transcript because it keeps me close to how I actually speak. Then I might use AI to pull the information together or show me where 1,200 words can become 900.

If it helps me say the same thing more clearly and saves somebody time, great. I still decide whether the suggestion sounds right, and plenty of the time the original language is better.

Whatever AI does along the way, the person whose name is on the work still has to make the call. That’s why I think judgement matters more as AI takes on a bigger share of the work.

When people become the premium option

I can already see it in content. Once it becomes cheap to produce something passable, a lot of it starts sounding the same. Work with a clear human voice stands out more.

I think customer service is going the same way. 87% of customers now say companies using generative AI for support still need to give them access to a human.

Klarna’s experience shows why that matters. Its AI assistant was doing the work of around 700 agents in 2024. A year later, its CEO admitted the cost focus had hurt quality. The company started hiring people again, with him describing human service as a VIP option.

When bots become standard, getting an actual person who understands the situation starts to feel more valuable.

I’m happy to automate aggressively, but I’m careful when the work depends on human judgement because that’s where cheap delegation tends to get expensive.

👤 Cory Blumenfeld is a five-time founder with two exits and the writer behind Taking My Time Back, where he writes about building businesses that don’t depend on the founder for everything. He’s built teams of 70+ and now focuses on helping founders win back their time through better systems, people and AI. Connect with Cory on LinkedIn.

Taking My Time Back

Helping founders, CEOs, and leaders take back 1,000+ hours a year with systems, teams, and AI 👊🏻

By Cory Blumenfeld

Through his agencies Redwerk and QAwerk, he sees what happens after the prototype looks convincing and real users start doing things you didn’t expect. Something that worked perfectly in a demo suddenly starts failing in ways you didn’t expect, from payments to duplicated bookings.

That’s why Konstantin sees AI as a multiplier. It can help you build more, but it can also lead you astray with a bad decision before you realise you’ve misunderstood what people actually need.

For founders, that shifts more of the work towards deciding what deserves to be built in the first place. Konstantin thinks talking to users matters even more now, especially when almost anyone can turn an idea into working software without spending months getting there.

In our conversation, Konstantin explains where vibe-coded products start to break, what still needs proper technical experience and how he thinks about building a software business when AI is available to everyone.

We also get into how he’s run the business over 21 years without outside funding, why he avoids hiring ahead of demand and what COVID and the war in Ukraine taught him about adapting when circumstances change underneath you.

🔗 Find Konstantin on LinkedIn

Key takeaways

1️⃣ AI makes bad decisions faster

Konstantin sees AI as an accelerator for whatever is already happening inside the business. If you understand the problem and know what users need, it helps you move further. If the thinking is weak, it helps you create more problems before you notice.

2️⃣ The demo can lie

Vibe coding can get you surprisingly far, especially for demos and early testing. Konstantin has seen products that look finished on screen while basic workflows fail once real users arrive.

3️⃣ Knowing what to build is the edge

When almost anyone can ship software, the bigger advantage comes from understanding your users well enough to know which problems are worth solving. More features don’t help if nobody needs them.

4️⃣ Your prototype still needs proper engineering

AI can move the point where you need a CTO or experienced developer much later. Once the product has real users and needs to work reliably, you still need people who understand the architecture and what’s happening underneath.

5️⃣ Don’t build for a business you don’t have yet

Konstantin has spent 21 years growing without outside funding and avoids hiring before the work is there. That same thinking applies to software: solve the problems you actually have instead of spending heavily on a future version of the business that may never arrive.

In this episode

00:00 Introduction to Konstantin Klyagin

01:59 DataForSEO sponsor message

03:21 More code doesn’t mean more progress

05:27 Build what users actually need

07:20 A prototype can look finished and still fail

10:45 AI multiplies bad decisions too

13:02 When you still need a CTO

17:52 Don’t hire ahead of demand

21:48 Understanding the customer becomes the edge

25:51 DataForSEO sponsor message

26:06 AI doesn’t belong everywhere