https://t.co/aX6JF1IyIh
Most AI memory research optimizes retrieval. But the bigger bottleneck is often memory writing. If the right information never gets preserved, retrieval can't bring it back.
Long-horizon agents don't just lose facts—they lose goals, decisions, constraints, and state.
That's exactly what we're solving next. - Coming soon.
Agents are producing increasingly more workflows & memories than humans now. What's your biggest pain deploying agents at work space? Comment if you face other pain points using agents as your coworkers.
@bittingthembits The "agents improving agents" loop is the right problem, but the hard part isn't the training ground — it's the eval. Whoever defines what "better" means owns the whole feedback flywheel.
@btaylor The "agent as the new UI" framing is the part most CX vendors will skip past — but it inverts who owns the customer relationship. When the conversation is the interface, the system of record becomes secondary to whoever holds the orchestration layer.
@VibeMarketer_ "Managing a fleet" is the real UX shift — the kanban metaphor barely survives once cards start spawning subtasks on their own. Curious whether dependency chains hold up when one agent's diff invalidates another's assumptions mid-flight.
@latentspacepod@pmarca@a16z The "almost here for 80 years" line is the part most people will misread as victory-lap. The interesting question is which of the four — reasoning, coding, agents, self-improvement — is doing the most work right now, and which is still riding on the others.
When I built the first Agent Marketplace MindOS in 2022 I thought agents will replace softwares. Now turns out agents = new labor, skills = new software.
Huge fan of your insights @lennysan@garrytan!
Inspired by your frameworks, I backtested 49 YC alumni histories to build a predictive skills leaderboard. Found some non-obvious patterns in what actually scales.
49 companies done, more skills generating. Check the Alpha: https://t.co/a9b8UiZQmG
Credits: https://t.co/jSTUdT95nL
https://t.co/GynQK0aBSC
Old SaaS = data + interface. But now they = data, readable by agents. Traditional SaaS will still exist but they will transform to data storage as source of truth for agents. New SaaS will transform into agents to replace labors (service as software). Next: Judgment & Supervision
🚀 BIG NEWS! 🚀 We just dropped powerful new features to help you work smarter:
🤝 Zoom Integration: Sync schedules & manage meetings in one place.
🎨 AI Posters: Create professional banners in seconds.
✨ Sleek New UI: Enjoy a smoother, faster experience.
https://t.co/7QarOytgpy
SF Tech Week! 🚀
Join founders, investors & B2B leaders to explore how Agentic AI is revolutionizing business. Let's talk fundraising, automation, and scaling.
🗓️ Oct 9, 3 PM 📍 Union Square, SF
Save your spot: https://t.co/jBnbsurtHl
#SFTechWeek#AI#B2B#AgenticAI
https://t.co/vkBoT0HnGe
AI’s $10T opportunity will be unlocked by memory.
Trends Sequoia sees:
- Persistent memory
- Seamless AI-AI protocols
- AI voice
- AI security
- Open source at a crossroads
That’s why we built Tanka → org-wide memory that persists, compounds, & grows.
Five Investment Themes (Next 12–18 Months)
5. Open source at crossroads
2 years ago: strong contender vs giants.
Now: precarious but essential for a free/open AI ecosystem.
Sequoia supports ensuring open-source remains state-of-the-art.