Sunday, August 30, 2026
The era of persistent AI coworkers has arrived
August 30 · 3 videos
OpenAI is moving past chat.
Tara Seshan says build for the 3 month horizon.
Google DeepMind launched Nano Banana 2 Lite.
Human preference is a lossy metric.
Doug Leone says fear of irrelevance is a tailwind.
The model is becoming the coworker.
“You fail if you build for where the models are now. You fail if you build for where you think the models will be in a year. Both outcomes are equally wrong. Only way to build is 2 to 3 months.”
SOTA Generative Media Panel — Dumitru Erhan, Shane Gu & Nicole Brichtova, Google DeepMind
Dumitru Erhan · AI Engineer · 56 min
Watch on YouTube →Google DeepMind researchers discuss the launch of Nano Banana 2 Lite and the strategic shift toward multimodal video APIs. They explore why language is a lossy pipe for sensory data.
- Nano Banana 2 Lite achieves 3 second latency for generative media.
- Gemini Omni Flash APIs are priced at parity with Y31 Fast to drive developer adoption.
- Language acts as a lossy pipe for sensory data like audio and skin tone.
- Users often prefer AI video for its sharpness even when it lacks physical realism.
- Internal testing missed artifacts like wedding rings appearing on hands until external testers intervened.
- Reducing generation time to 3 seconds transforms models from iteration tools to production tools.
AI’s third era: the rise of persistent AI coworkers | Tara Seshan (OpenAI’s product lead)
Tara Seshan · Lenny's Podcast · 81 min
Watch on YouTube →OpenAI Product Lead Tara Seshan outlines the transition from ephemeral chat to persistent AI coworkers. She explains why the 2 to 3 month horizon is the only viable product strategy.
- Knowledge workers are shifting from executing tasks to steering AI with intuition.
- Building for model capabilities more than 3 months out is likely to fail.
- Organizations should prioritize mocks over docs to shorten feedback loops.
- The new bottleneck for progress is human ambition rather than technical capability.
- OpenAI uses a founders led model where leads act with high autonomy.
- Share strategy documents at 70 percent completion to invite collaborative polishing.
How to Dominate for Decades | Doug Leone, Sequoia Capital
Doug Leone · David Senra · 82 min
Watch on YouTube →Former Sequoia Capital leader Doug Leone discusses the AI revolution as a fundamental rewriting of business. He shares heuristics for identifying outlier founders and the cost of selling winners too early.
- Leone views himself as a low level analyst to learn the AI domain from scratch.
- Venture capital success requires finding hypercompetitive founders with 150 plus IQs.
- Sequoia once owned 20 percent of NVIDIA but notes the high cost of selling too early.
- The trust equation is the essential accelerant for high performing organizations.
- Maintain a sniffer for reality by living without personal assistants or handlers.
- Investment bets must have the potential to return the entire fund.
References
PeopleDumitru Erhan (@doomie) · Shane Gu (@shaneguML) · Nicole Brichtova (@nbrichtova) · Tara Seshan · Doug Leone · Michael Moritz (https://sequoiacap.com/people/michael-moritz) · Jensen Huang · Patrick Collison · Andrew Ambrosino (@ajambrosino) · David Vélez · Don Valentine · Fred Luddy · Logan Kilpatrick · Jitendra Malik · Dylan Field · Ari Weinstein · Mike Speiser
ToolsNano Banana 2 Lite · Gemini Omni Flash · ChatGPT · Codex · Stripe · Nubank · Apple · Cisco · Google · NVIDIA