Anthropic runs a small team for its size and filters aggressively on one thing: does this person care about the mission more than the money.
Lucas Swisher's (@LucasSwisher1) point is that every AI lab can offer life-changing comp right now. That's table stakes. So when researchers pick between OpenAI, Anthropic, and Google, the decision isn't happening at the compensation layer anymore.
It's happening at "do I get to work on the frontier, on products I believe in, with impact I can see."
An incremental $10M over three years just doesn't move that calculus for the people actually worth retaining.
2
123
Manager's time is easy. Meetings just happen to you. You show up, you react, you're done.
Maker's time requires you to activate yourself. And that's exactly why it keeps losing.
Patrick Ewers' (@PatrickEwers) fix: have your EA track how many hours of protected thinking time you actually used each week. Four out of six? He calls that a win.
Then before you sit down to think, calm your brain, get it into a genuinely good mood, and define the single first step. No ambiguity, no menu of options.
Ambiguity is what turns two hours of deep work into "let me just answer a few emails."
14
Ten back-to-back Zoom meetings can still be a wasted day if none of them are memorable.
Patrick Ewers' (@PatrickEwers) fix is simple math: make your one-hour meetings fifty minutes. Nothing you need to get done in sixty minutes can't get done in fifty.
That gives you five minutes to debrief and send the follow-up, five minutes to prep for what's next.
Same ten meetings. Actually effective.
1
176
Anthropic has lost one of its original 17 employees. That's it.
Lucas Swisher's (@LucasSwisher1) point: in AI, talent is the moat in a way it never was in SaaS. Every SaaS company "tasted like chicken." Same playbook, same features, same outcomes.
AI punishes that sameness. Look at Databricks. Ali has taken that company through three completely different eras, moving data, becoming the warehouse, now the AI data layer, without losing the thread. Snowflake had the same shot and is growing at 34% while Databricks is past 80%.
Compute is table stakes. The team that can reinvent itself every cycle is the actual weapon.
3
130
Everyone building an AI application right now is racing to $100M ARR as fast as possible.
Lucas Swisher (@LucasSwisher1), Co-Head of Growth Investing at Coatue, points to Harvey as the counterexample. Unique data deals. Embedded networks between law firms and corporates. A harness built to use every tool, including open source.
"Being number one in the category really, really matters. The disproportionate share of the gains in technology come from being the number one."
Growth is downstream of those early decisions, not a substitute for them. Skip the moat and you end up as one of 50, with a leaky bucket and no real advantage when the model labs decide to come for your space.
3
636
Lucas Swisher (@LucasSwisher1), Co-Head of Growth Investing at Coatue, thinks the Mag Seven is already dead as a concept.
His evidence: the Nasdaq is up 18% over six months, and not a single Mag Seven stock has matched that. If the bundle isn't beating the index anymore, it's not really a bundle.
His replacement: an AI Eight. SpaceX, OpenAI, Anthropic, ByteDance, Amazon, Google, Nvidia, and maybe TSMC.
Worth noting ByteDance alone is closing in on $100B in net income, still growing 30% a year. The growth moved. The label didn't catch up.
12
7,850
Leo Polovets (@lpolovets), GP at Humba Ventures and co-founder of Susa:
"Software is pretty scarce before, but now with AI, it feels abundant."
Cheaper software means more competitors shipping the same thing faster. Polovets thinks the value migrates somewhere else next: hardware, deep tech, the stuff you can't spin up with a few prompts.
His bigger realization after a decade plus investing: moats matter less than people think if you're fighting for space in a crowded lane. A great product in a crowded market still struggles to stand out.
"I've just become more and more of a fan of companies doing unique things, like funds with very unique approaches. I think that's the best and most predictable way to win."
Own the lane. Don't just defend it.
2
238
Executive coach Patrick Ewers (@PatrickEwers) on why AI drafts actually get emails sent, not just written:
A half-decent AI draft turns a leader from a reluctant author into an editor. Editing has no ego attached. Writing does.
But it's still their content. They dictated it. The AI just removes the friction of staring at a blank page, not the ownership.
And the tell that it worked: when the words you actually dictated survive the edit and show up in the final send.
3
165
Michael Green (@profplum99) calls it what it is: Citadel and Jane Street aren't just market makers, they're facilitators who profit by seeing the crowd before the crowd sees itself.
"They take the passions of the crowd and say it's far more profitable for me to facilitate their insanity, effectively front running their trade."
That's why Citadel bought Robinhood's order flow. Same reason Jane Street chases lead market maker status in ETFs. Visibility into the order flow is the edge.
Green's line for it: they've become the croupiers at the casino.
1
51
A founder emails an investor with an "emergency," needs to talk in 24 hours.
Predictable response: yes, always. Unpredictable part: the investor is in back-to-back meetings for the next three hours and can't say it himself.
Patrick Ewers (@PatrickEwers) on inbox shadowing: the EA sees the email is red hot and replies from their own inbox. "David asked me to get him on the calendar today. Does 6:45 work?"
Founder hears back in ten minutes instead of five hours.
Most GMs claim they'll roll up their sleeves for you. Almost none can actually deliver that response time. The ones who can just built the system for it.
1
2
217
