Everything is a wrapper

Excited to be a small supporter of Trajectory! Today, most production agents for serious enterprise use cases require a deterministic harness that encodes edge cases or company-specific business processes. In this paradigm, to make agents better, companies make their deterministic tree / policy graph bigger and bigger since models can’t learn on their own. Trajectory re-writes this paradigm. Their platform helps models learn from a company’s own product usage data, building the rails for specialized AGI. And they hiring fast, DM if interested!

Excited to partner with Hilmar and the Fenris Games team as they become an independent game studio. And pumped for their new partnership with @GoogleDeepMind, who will be investing with us. As someone who’s played thousands of lifetime hours of strategy games, this has been special!

Similar to looking at lines of code written, token spend isn’t necessarily indicative of productivity, but if the number is 0, you’re probably doing something wrong

Appreciate the way Neo helps high slope young teams. Excited to meet the new group!

FCF continues to elude large model companies because: -The threat of distillation forces Anthropic to keep running on the hamster wheel and disrupting themselves in short cycles (FCF comes the longer they are able to harvest the fruits of their previous model before they have to invest in the next one) -The move to RL democratizes the frontier, which was previously a capital intensive game Companies that own the user (the application layer) are best positioned since their usage data can be used for RL in this new paradigm At the same time, while their LLM API may not be profitable, it may be a low margin loss leader to sell other high margin software products (Claude Code/Codex, enterprise deployments, etc). This is the playbook profitable supposedly “commodity” API businesses like Stripe and ElevenLabs have leveraged

Fascinating data on the challenges growth areas of gaming: It's harder than ever for new games to compete, with older/entrenched games holding or growing share of playtime across all platforms Roblox dominates with kids and accounted for 2/3 of the entire industry's net new non-China growth... …but once kids get old enough, spending shifts to social media and gambling (iGaming, sports betting, prediction markets, crypto)

Congrats to @elevenlabs on their $500M fundraise. @matiii has a rare combination of humility and obsessiveness, and it sets a culture that attracts and retains the best. It’s why they’ve shipped product at lightspeed even as they’ve grown and have consistently beat even our “optimistic case” when we first invested in the Series C in 2024. Congrats team!

Privates markets love a non-profit with a strong consumer brand like OpenAI and Utah: Some ideas for what’s next: -The College Board (~$1B revenue) -Goodwill ($6B revenue) -Habitat for Humanity ($350M revenue)

Complaints against VC king-making miss the whole point. It’s not about VC $ crowning the winner. The market is pricing in the value of market leadership. For a long time, "value investing" (backing the #2 player at a cheaper price) has lost to paying high prices for the winner. This logic taken to the extreme means investing in companies ahead of knowing if they'll be market leaders since the prize is so great if you're right.

Calling BS here 2 reasons: -Just because a product looks good in a demo doesn’t mean it performs well in a real-world environment with the required integrations -Industry-specific GTM matters a lot. And winning the customer with a useful but not defensible agent can be a wedge to something bigger

We've raised $7M to help companies build AI agents that actually learn and work. @Osmosis_AI is a platform for companies to fine-tune models that outperform foundation models with reinforcement learning. Better, faster, and cheaper.

Most AI generated videos looks fake, but that’s the wrong question to ask since the world only cares about the top 1% of content anyways