Sep 18, 2025 · 35m · latent-space
⚡️No, Don't Do Palantir for AI - Brendan Falk, Hercules (AUDIO FIXED)
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Former Fig founder Brendan Falk reflects on his journey attempting to build an 'AI-native Palantir' with Zeus, breaking down the operational and economic challenges of enterprise AI consulting that ultimately led to his pivot toward AI app generator Hercules.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 13.3% of the talking time here. How this is scored →
The hosts as informed peer 2.7 Guest teaching 4.6 Guest disagreement 1.4 The hosts pushing back 1.3
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 32:25 Brendan rejects the engineer-bubble premise
Brendan directly counters Swyx's point about model quality perception by emphasizing that outside the engineering echo chamber, the general population cares only about functionality rather than which underlying LLM is served.
Hardest push from the hosts ▶ 32:14 Swyx challenges hidden margin optimization via model degradationSwyx pushes back against Brendan's margin thesis by citing community backlash whenever users suspect providers are stealthily serving quantized or degraded model versions.
Biggest teaching moment ▶ 17:05 Brendan details real-world enterprise edge casesBrendan dismantles theoretical AI automation assumptions with real-world examples, explaining how bizarre enterprise workflows like Excel sheets embedded inside Word files derail end-to-end automation.
The host holds their own ▶ 29:45 Swyx connects AI app generation to Karpathy's Menugen thesisSwyx demonstrates domain expertise by connecting the competitive landscape of Hercules, Bolt, and Lovable directly to Andrej Karpathy's Menugen framework and Cursor's underlying margin constraints.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Discovering Enterprise AI Reality at Amazon Private Equity | 1 | 4 | 1 | 0 | Brendan delivers an opening monologue describing his experience as AWS AI lead for private equity portfolio companies. The hosts let him speak uninterrupted as he contrasts Silicon Valley's AI optimism with traditional enterprises that barely knew what Anthropic was. | |
| The Original Thesis of Zeus: Building Palantir for AI | 2 | 5 | 1 | 1 | Alessio asks why anyone would attempt to build a Palantir for non-technical enterprises. Brendan details how traditional system integrators like Accenture run endless non-production POCs, framing the original thesis for Zeus targeting high-value enterprise contracts. | |
| Contract Economics and Enterprise Go-To-Market Challenges | 3 | 6 | 1 | 2 | Alessio queries why Brendan did not just distribute through PE portfolio companies. Brendan explains the contract economics, illustrating why $500k contracts take equal effort to $5M contracts and why custom workflows require $5B+ revenue clients to make economic sense. | |
| The Breakdown: Messy Data, Maintenance, and Existential Competition | 3 | 7 | 2 | 1 | Brendan walks through the breakdown of the business model, explaining unmaintainable edge cases (such as Excel sheets inside Word files), why MCP does not fix messy internal data, and the existential threat posed by well-capitalized vertical players like Decagon and Sierra. | |
| The Pivot to Hercules: Returning to Developer Tools | 3 | 4 | 1 | 1 | Alessio asks about the pivot to Hercules and why they chose a web-first app builder over CLI tools given Brendan's Fig background. Brendan reveals his two-year non-compete with AWS expires in twelve days, teasing future moves. | |
| The AI App Generation Landscape and Software Economics | 5 | 4 | 3 | 4 | Swyx brings up Cursor's gross margin issues and Karpathy's Menugen thesis, pushing back on the assumption that app platforms can easily swap in cheaper, quantized models without users noticing. Brendan counters that mainstream SMB users do not behave like engineering power users. | |
| Reflections on Startup Transparency and Final Takeaways | 2 | 2 | 1 | 0 | Swyx thanks Brendan for publicly documenting his startup's pivot. The conversation wraps up with reflections on Silicon Valley hype culture versus the benefits of transparent founder post-mortems. |
Statements from this episode (11)
AWS has about 20,000 private equity-backed customers
“And what's interesting is Amazon has about AWS has about 20,000 customers that are backed by PE firms.”
Falk: Zeus originally targeted Global 2000 companies for $5M-$10M contracts
“And so ended up leaving Amazon, starting this company, Zeus, And, you know, the idea, pretty simple. Number one, we would only go after the global 2000. That was very, very deliberate. Two is we are, we were building very custom software. So the idea was we wa…”
Falk: Zeus pivoted away from enterprise AI transformations within six months
“And that's that was the key pitch that we started with. And obviously we pivoted away from it about six months later.”
Falk: Scoping large custom enterprise contracts requires only marginally more work than small ones
“It's the amount of work for like a 500 K contract versus a million dollar contract versus a five million dollar contract. It's not proportional. It is like, you know, a little bit more work to get a much larger contract. And so it's just far more optimal to go…”
Falk: AI cannot solve enterprise messy data because it is a people problem
“I could see a path to AI solving the integrations problem. I really couldn't see a path to AI solving this sort of messy data problem because it's more of a people process problem than it is like an AI software problem.”
Falk: Decagon and Sierra will always outperform generalist enterprise AI builders
“I can just guarantee that Decagon is always going to be better. Sierra is always going to be better because they spend all day, every day thinking about it.”
Falk: The Palantir deployment model fails in AI due to enterprise churn
“This sort of Palantir model really only works if you not, you do a lot of like, you have very low or even negative gross margins in the first year, but then once it's deployed into production, you take the head count off, the margins go up and you have this ni…”
Falk: Enterprise AI services only work at $5M-$10M scale or single-vertical focus
“One is you could keep doing the model, but only go after projects that go up to ten million dollars. I think that's really the only way this actually works or five million dollars, not even at the company level. It has to be the one project that can scale up t…”
AWS imposed a two-year non-compete clause during its Fig acquisition
“When we got acquired by AWS, we actually had a non-compete for our product for two years. And so I've got 12 days until that non-compete expires.”
Falk: ChatGPT will aggressively enter and replace design prototyping tools
“And candidly, I think that ChatGPT is going to come in and replace a lot of that because that's what they did in a GBT five demo. They were really showing that off. These things are not designed for production. It's just another reason to come to ChatGPT. So I…”
Falk: Consumer AI app builders have better margin potential than developer CLIs
“And so I think that is what enables us to ultimately see a path to actually pretty high gross margins. Whereas the CLI based and the sort of engineering focused ones, I think you're in a tough position because everyone always wants the best model.”