Joel Salinas from the Leadership in Change newsletter invited me onto his Substack Live, and we ended up having a useful conversation about where founders still use AI too loosely at work.
We got into ChatGPT vs Claude, role-based AI setups, why weak output is often a setup problem, and what changes when you stop treating one tool like it should do every job.
A lighter one than the usual Millennial Masters podcast, but plenty in here if you’re trying to make AI more useful inside real work.👇🏻
What came up in the live
1️⃣ Stop using one chat for everything
A lot of weak AI use starts with people treating one chat like it should hold every task, tone, and context at once. Once the work gets more varied, that setup starts slipping fast. ⏱️ 08:19
2️⃣ One AI tool shouldn’t be doing five different jobs
Research, writing, editing, analysis, strategy, and admin all need different standards. A lot of disappointment starts when one tool gets asked to do too many of them badly. ⏱️ 11:33
3️⃣ If the output is weak, look at the setup first
Examples, tone, banned words, context, previous materials, and clear standards all matter more than you think. A lot of what gets blamed on the model is really bad setup upstream. ⏱️ 12:19
4️⃣ Voice dictation is still badly underused
If you’re still typing compressed prompts, you’re often starving the model of tone, detail, and intent. Speaking your thoughts out properly gives it much better raw material to work with, and it’s much faster than typing. ⏱️ 16:46
5️⃣ A better setup beats a clever prompt
Once the conversation moved into orchestrators, sub-agents, and repeatable workflows, the bigger point was that useful AI is less about clever prompts and more about building a better system around the work. ⏱️ 19:04
6️⃣ ChatGPT vs Claude matters less than you think
The useful question is not which one “wins” in the abstract, but which one fits the job, the workflow, and the level of context you need. Tool choice matters, but setup and role clarity matter more. ⏱️ 25:00
7️⃣ You still need to decide what stays human
Delegation matters, but so does judgement. AI can take work off your plate without taking the parts of the job that still need your taste, voice, and final call. ⏱️ 52:24
8️⃣ Good AI use looks more like management
Once AI becomes part of the workflow, the founder job starts looking less like “using a tool” and more like setting roles, standards, and feedback loops properly. ⏱️ 53:44
A few of the ideas we touched on came from these pieces:
Joel’s features: Start building AI teams | How to use Claude Cowork task scheduler | 5 Claude skills that save me 40 hrs | Cross-domain thinking as an AI skill
Millennial Masters articles: You’re the AI manager now | Managing AI like a bad hire | What to delegate to AI without losing control | How not to sound like ChatGPT
