Check how many selling windows exist per year
When evaluating a market, check how many selling windows exist per year — ed tech only allows selling to schools twice a year, which caps growth.
described
When evaluating a market, check how many selling windows exist per year — ed tech only allows selling to schools twice a year, which caps growth.
described
Increase the frequency of a price-monitoring job to match how often the underlying prices change — daily if the wholesaler updates prices daily.
endorsed
When deciding whether to pivot, honestly weigh the counterfactual — 'maybe we would have figured a way' — but recognize that without a breakthrough idea in your brain, resourcefulness alone won't make it a breakthrough business.
described
Randy Hunt recommends framing team success as shared — 'When we do great work, we're all going to win together' — selling that vision, and then getting to work building.
Randy Hunt observes that doing the work yourself, even imperfectly, builds appreciation for the craft of serious professionals — designers hand-coding front ends came to respect their engineering partners — and that this shows up in healthy cultures where craft is respected.
The narrator discourages treating design reviews as debates to win with sharp rationale, recounting that this pattern kept losing to leaders who simply didn't like the direction.
Sujee recommends using large teacher models to train smaller student models, noting Nebius hosts both large and small models, and that large 'teacher' models are very capable and smart but also large in size.
Alex Lieberman argues that spending all day in the guts of enterprises reveals enormous nuance, inertia, systems, and people that vertical AI solutions must wrangle. Buyers now expect one-of-one software solutions, driving every software company to build teams of forward-deployed engineers, and vertical AI companies face the question of how to deliver one-of-one experiences at software margins. The moat for vertical AI companies like Harvey is access to data that ChatGPT will not have; Anthropic and OpenAI are acquiring the same data sources but are not as dedicated. Harvey has Tabular Review and Legora has similar affordances — customized UIs because they only serve their vertical — and the pitch for vertical AI is focus, caring about the last mile the way you care about yours. He uses a
some claims are hypothetical or attributed
CJ Hess endorses the principle that the value is sitting down with the customer, hearing everything they can brain dump, and providing that to the model to get closer to the solution.
Data companies like Micro One, Mercor, and Handshake get 90+ percent of their revenue from the labs. A number of these founders argue it's going to flip-flop to enterprise-driven in the next few years. The host is skeptical: if enterprises become the core buyer, they'd be fine-tuning their own models — a hard bet because the bar keeps moving on the frontier, and token costs keep going down.
transcript garbled; company names uncertain; founders' prediction, not fact
The speakers describe how token costs continue to go down on these models and costs are coming down on a per-task basis, hitting a Pareto point. Six to eight months ago was the point at which you would be really concerned about token maxing costs; today it is way harder for someone to token max and really hurt the company in terms of cost. There was an era where everyone was using Astra and Fable because they gave the best results, but those models are expensive; in a world where Astra 6.5 and Fable 5.5 are super expensive but get the best results, they could see trending back that direction. Muse Wonder launched yesterday, is on the frontier, and is dirt cheap. Competition is heating up in both open source and closed source, which will continue to drive costs down. A 10x smarter model, on
A Jevons Paradox of Privacy likely exists for the assistant market — making AI email reading cheap means people allow far more email reading than they would with humans. The speaker hedges: 'it feels like that probably exists.'
speaker hedges; framed as probable
The speaker's framing: get humans to operate at the speed of AI, not at the speed of humans. The company believes different things about the relationship between a human and AI, and is probably more forward-thinking than even the labs.
speaker's framing
Part of being enterprise-ready is not looking at user data, so the company instead looks at aggregate metrics of who their power users are on the platform.
described with 0.8 confidence
Most AI products today will have very similar capabilities — most harnesses in the work context will be able to do the same things. Product features that are ahead of market — the traditional Silicon Valley differentiator — will be more and more difficult to sustain as an advantage.
speaker's assessment
The speaker thinks of the market as building a product that works with people to find more and more opportunities for them to deploy AI to be better at their job, and to offer them to the user.
speaker's framing
The company's belief is that custom software all the way through can be built using AI — offering 80-20 very customized software for people over time.
company belief
The speaker argues that Amazon, having been the aggregator of supply, may not have a good business model in a world of AI agents, and that there is little it can do about it.
framed as a prediction/interpretation, not a certainty
The market for executive assistants seems small because most people don't have one, but if produced via silicon the market is quite large.
interpretation
Randy Hunt observes that AI research agents dramatically reduce ineffective research — research done well methodologically but that never impacts the product — because light insights no longer require heavy processes to surface.
No published protocol matches this selection.
The creators behind most of what cleared.
195 cleared insights match
99 cleared insights match
85 cleared insights match
Choose sources → to get a library of your own and a daily brief.
