Panel prep · Carlile Advisors
How to actually use this page.
This page is for browsing and listening, not for preparing. Reading research feels like prep and is not. The preparation is saying your answers out loud and rewriting them in your own words. Use this to explore what the experts said and to pick what to listen to. Do the real work in the editable doc, and walk in on the 14th with a one-page card.
Empathetic and compassionate, with a little optimism. Practically that means one thing: name the hard part before the hopeful part. That order is what makes optimism land instead of grate, and there are people in that room who were recently laid off.
One distinct point each. Tap to open. The failure mode is giving the same "be curious, use the tools" answer four times.
Stop studying AI. Pick one problem you already care about and use AI on it this week, then walk into the interview with that thing.
Lead with the empathy beat, before any advice:
Honestly, the first thing I would do is give myself some grace. This is a genuinely hard moment to be looking, and a lot of that is not about you.
Then the turn, then the fantasy football story in full.
What I would do differently is stop treating AI as a subject to study and start treating it as a tool to point at something I already care about.
Then the actually useful part:
In the interview, do not say "I'm familiar with AI." Everyone says that and it means nothing. Say "here is a thing I built, here is what it saved me, and here is the first thing I would try at your company." That is a completely different conversation, and almost nobody is having it.
You need to be able to read and judge code more than write it from scratch, because reviewing what the model wrote is now the job.
Less than you did three years ago, and understanding what is happening matters more than ever. Those sound contradictory. They are not.
You cannot approve what you cannot evaluate. The model will hand you something that runs. Whether it should ship is still a human call, and that call is the actual job now.
What to tell them to actually learn, because "learn architecture" is useless on its own: how data moves between systems and what an API is; front end versus back end, roughly; what good looks like versus what merely runs; how to debug. Skip memorizing syntax, that is the part that genuinely changed.
Close optimistic:
The barrier to your first working thing is lower than it has ever been. You could build something real this weekend that would have taken me a month five years ago.
Taste and follow-through. Same tools, different judgment.
Define taste out loud or it is a buzzword:
Taste just means knowing which of the five things the model gave you is the good one. That is it. And you only get it by doing the reps, which means the people who have actually been using these tools on real problems are ahead, regardless of age.
Then the unglamorous one nobody else will say:
The other thing is showing up when you said you would. That sounds like the most boring advice in the world. It has also never been more of a competitive advantage, because everybody's output looks polished now. What is scarce is the person who actually does the thing.
Problem framing. Knowing what question to ask.
Everyone is going to say communication, and they are right. The version I would add is knowing what question to ask. These models will confidently answer whatever you put in front of them. They will not tell you that you asked the wrong question. Figuring out what the actual problem is, before you ask, is the skill I see separating people right now.
Forty seconds, then stop. Hannah takes the people side, Chandresh will likely take judgment or ethics. Brevity here reads as confidence.
You do not keep up with all of it, and nobody does. Go deep on your own field's edge and let the model teach you the rest on demand.
Take the pressure off first. This question is really about anxiety:
Here is something I think would help to hear. Nobody is caught up. Not the people building this, not me, not anyone on this stage. The field genuinely moves faster than any person can track. So if you feel behind, that is not a personal failure, that is just the actual condition of the field right now.
You do not need to keep up with everything. You need to know the edge of your own field. One tool, one hour a week, applied to work you are actually doing.
Then the close. Slow down for it. Do not add anything after it:
And the part people forget. You can just ask it. If you are nervous, if you do not know what to learn, if you are embarrassed that you do not know where to start, you can type that exact sentence into one of these models and it will help you. It is a one on one tutor that has infinite patience and will never make you feel stupid for asking. For a lot of people in this room that is the single most valuable thing about this technology, and it costs nothing.
Not yours, but be ready. Hannah owns Q2 (biggest AI mistake), Q3 (ATS resumes) and Q5 (jobs redefined). Do not answer them unless invited. If Eshwar opens Q5 up, your 20-second add from real consulting work: what goes first is not whole jobs, it is the glue work inside them. The copying between systems, the reformatting, the status update nobody wanted to write. The job stays and the boring third of it goes.
Verified against real transcripts. Cap yourself at three quotes for the whole panel, or you will sound like you are reciting.
"It shifted to where AI is doing some basic autocomplete, maybe writing 5, 10, 15, 20% of code, to writing 80% of code."
Lex says 80. DHH answers: "Or 100."
"We let them vibe. And we ended up with a lot of PRs that individually perhaps could have been justified for a hot moment, but taken all together, destroyed the architecture of the system."
"Most organizations don't know what they want. They're not bottlenecked on implementation. They're bottlenecked on ideas. They're bottlenecked on vision. They're bottlenecked on taste."
"I don't care what field you're in, you should be playing with this stuff... the best way to protect against any risk of your career being obfuscated or eliminated from AI is to be the most AI-enabled version of yourself you can possibly be."
"In any field, one thing I love to suggest on the learning side is, know the history and know the new innovative edge. If you bring both of those things to the table, you are highly compelling."
"We don't build for the model of today. We build for the model six months from now... try to think about what is that frontier where the model is not very good at today, because it's going to get good at it."
"The AI is very good at following instructions, and it's probably going to be hard for humans to compete with the AI on just following instructions reliably... it's going to require: how do you know to do things yourself and how do you have agency and independence?"
Ranked. If you only get through one, make it the first.
One to check, not to trust. All-In episode 275, May 29, 2026, is the likely origin of the "this generation grew up with AI" line you remembered, but the finding is that your memory blends Gurley with Sacks and Chamath. Nobody has confirmed it against the full transcript. If you want that framing, open it yourself first. Otherwise skip it, the Gurley material above is better sourced and better for this room.
Two are worth memorizing. The rest are backup if challenged.
Open gap. The research on entry-level hiring and over-40 displacement was lost in a handoff failure and never landed. That is the evidence that would let you speak honestly to the hardest people in that room rather than around them. Worth re-running before the 14th if you want to go there.
Each of these is a real way to lose the room or lose credibility with the professor sitting next to you.
The goal is not to know more. It is to have said it before.
Who is on stage and who is in the seats.
Loose end. The deck's question-ownership slide spells your name "Tucker Carlisle." Worth one line to Eshwar when you reply.