1/ In 2025, you’d think builders would stop defaulting to “let’s train.” Scaling laws are real, ofc. ++ model power keeps raising the ceiling. But capability ≠ adoption. The moat in workflow AI is bottling institutional context and making it useful in the moment of work. 2/ Quinn Emanuel strikes fear. It didn’t get there with clever clauses alone. It’s decades of scar tissue: judge-specific heuristics, forum strategies, discovery playbooks, damages theories tested in high-stakes trials. That accumulated edge is institutional capital.

Was thinking about early warning signs things are going sideways in startups today. “Code smells” aren’t bugs per se, they’re early warnings and startups have them too. Some that make me nervous
1/ The “god object” in code, one class that knows everything, shows up as the founder who must approve everything. Early on this protects taste. Later it throttles throughput. The trick is to export your “taste” into principles, examples, and defaults so others can act like you 2/

1/ early thinking about a side project:

given advances in AI, there should soon be a very easy way to find and create kids educational content that is personalized to interest, reading level, and interactivity/modality 2/ for example, if you have a 7yo interested in deep sea creatures or designing games or mythology, how can you hit the right point of the skill/learning curve? high quality specific content is rare (books probably worse than what you could ai-generate at this point) and Amazon search is garbage

fun to do a last minute “state of ai startups” talk (w/ @prnvrdy) this am at #neurips for old friend @swyx at his live paper-reading club

couple quick takeaways below on what’s working, where we disagree with the dominant narratives, and some of our latent priors in investing👇
AI years are like dog years: what’s changed

1/ As AI models grow more powerful, it’s becoming clear that data strategy is just as important as model architecture for advancing capabilities. The right data—and how we use it—will determine the future of AI performance. 2/ I was explaining to someone (not in AI) that neural networks don’t just “memorize” word sequences. Instead, they learn deeper patterns, allowing them to predict the next word even when they haven’t encountered that exact combination before..

Let’s talk about runway 🧵

1/ “Runway” is really a cursed way to think about startups and the remnant disease of the 2017-2022 VC bubble

If cash out the door is investment, it must have ROI, eventual translation into free cash flow (FCF) 💰 🌊

If cash out the door has no… twitter.com/i/web/status/1… 2/ Some companies and products obviously require more investment than others, but clearly not all investment in technology is equally commercially valuable. One can engineer something very hard to build that no one cares to pay for. It’s quite common

1/ Incredibly easy to make something cool with LLMs, but hard to make something robust and consumable by end users. 2/ We’re seeing lots of aggressive seed/A venture investing is happening right now in sexy recorded demos (but most are like movie sets) and early developer interest (but most tools can’t be used for anything in production at scale).

1/ Excited for Harvey.ai (AI Superpowers for Lawyers) and their $20M Series A fundraise led by @gradypb @sequoia. We at @w_conviction doubled down, as did other seed investors @OpenAI @eladgil. Read more from founders @gabepereyra @winstonweinberg… twitter.com/i/web/status/1… 2/ When I first met the founders, I was struck by the rare combination of @gabepereyra’s experience working on the state of the art in AI at 3 top labs (Meta’s FAIR, Google Brain and DeepMind) with customer-centric pragmatism, and a clear-minded focus on the trust necessary to… twitter.com/i/web/status/1…

1/ the funny thing about the parameter-maximalists who argue for “one model to rule them all, just stuff all your facts, context into the prompt, assume unlimited context length coming” is… 2/ 32K (or even magnitude larger) is pretty far away from “unlimited context length.” I am sure developers will continue bumping up against this ceiling

1/ founders are going $0-10M in ARR first 18 months of (efficient) selling, leveraging LLMs against “niche” use cases 2/ despite the AI hype, there’s plenty of real customer value, retention and repeat usage

1/ EOY prediction time! 2023 will be the year of racing to incorporate AI in products, but most companies' first instinct (shallow integration of generic models available through limitied APIs) is only table stakes.

The race is now on to: 2/ drive down the cost of pretraining, ongoing training and inference (still prohibitive in many use cases)

1/ 🧵Everyone who thinks VCs are cold-blooded has never met an activist public markets investor. I worked on “activism defense” for tech companies during my tenure at Goldman. #StarboardValue took stakes in $WIX, $SPLK & $CRM this week - quick take below:

bit.ly/3FxtMaN 2/ They see an opportunity to buy strong growth at less premium, as well as reap the cash flow benefits of increased focus on profitability

Some personal news: We raised $100M for @w_conviction, an new early-stage venture capital firm purpose-built for Software 3.0

sarahguo.com/blog/conviction 1/4 No soundcloud, but: if you are a founder or investor and want to shape the future of AI, please reach out! We are growing our team and always on the hunt for people starting enduring companies. sarah at conviction (dot) com

1/ Right now, founders are despairing that they can't possibly change their business *enough* within the 2+ year runway they have. But this is based on the (skewed) high-cash, high-valuation but slow-action norms of the 2015-2021 startup environment. 2/ I just got back from dinner with @dugsong (and a bunch of other people I admire - thanks @ericwu01) who only ever spent $14M to build @duosec to $100M+ in ARR before its sale to @Cisco.

1/ PSA 🚨 from your local cybersecurity and safety nerd!

Have been experiencing a pretty sophisticated social engineering attempt recently. Thread on what to watch out for: 2/ Step one: Someone attempts to reset my @Apple password from a new device in Sacramento (I’m not there)

1/ The assumption of some growth investors that they’re going to permanently get paid billions for just identifying companies that are somewhere along a “triple, triple, double, double, double” ARR trajectory seems sus 2/ Markets are not that inefficient. If all you do is look at two numbers like ”3X ARR” and 110%+ NRR” you don’t have a differentiated strategy

1/ I was today years old when I was advised for the 2nd time in my career to “be less stylish” because l might “intimidate engineers” I am working with. 🤯 2/ Five years ago I suppressed my feelings and nodded blankly when told this, but this time I just blew up. What the hell. Reinforcing Silicon Valley brotopia is not helpful.

1/ 🧵Discussion from this AM: Good early stage SaaS startup engineering today is a game of intelligently walking the tightrope of shipping speed and scalability/quality. Most fall off on the wrong side (the latter). 2/ Teams need to A) stand up end-to-end systems in order to learn from users as quickly as possible and B) rapidly rebuild, rearchitect, deepen parts of the product as they learn about requirements. Who is responsible for those two opposing forces in your company?

1/n Thinking about how platform companies get built. Some startups (@stripe) have massive, secular TAM that they can just drive a truck through for a decade or two -- even if it's not obvious day 1. Awesome -- this is the minority. 2/n For most companies, market size is the wrong question at the Series A.

What a mistake it is to title-chase when choosing a startup 🤦🏻‍♀️.

SVP at a company goes nowhere is SVP of nothing. Early teammate at a company that becomes a $10B+ platform biz, grows into a staff engineer, an SVP of product, an extraordinary leader.

Choose company, not title. 1/n I've had a number of people (@MinaMarkham @polotek @efr_schultz and more) point out this view lacks nuance, or add good points. I hope I am hearing them correctly - let me add some context and adjust👇🏼:

#TuesdayThread on B2B Product👇🏽

1/ talking to a friend this morning who went from consumer product gal > enterprise founder about why "consumer-grade product" is a concept 2/ as an enterprise product person, this kind of offends me. but! there is a grain (maybe an entire bag) of truth in this criticism. I've given entire presentations about why is enterprise product overall not as strong on usability (and why that's a huge opportunity)