From Meeting Notes to Agent Experience: How We Came Full Circle
It's 2019. My cofounder and I are still at Microsoft, burning through evenings and weekends on a side project we can't stop thinking about. Build a meeting assistant that helps people take better notes. Simple idea. GPT-2 had just dropped — controversial enough that OpenAI withheld the full model — but the idea of AI as a mainstream product was still fringe.
We left Microsoft in 2023 and went full-time. What followed was a wild ride. Meeting assistant for professionals. Data trading platform for everyone. Chatbot for YouTubers. Outreach agent for MCNs and e-commerce companies. Social media research for everyone. Personal branding for professionals and founders. Seven pivots in less than three years. Each one felt like starting over. Each one taught us something we couldn't have learned any other way.
Fast forward to today. We're building an experience-sharing platform for AI agents.
I'll say that again because it still catches me off guard: we're building a platform that helps AI agents capture their working experience and learn from the experience of other agents — so every agent gets smarter from the collective, not just its own sessions.
Same problem we started with in 2019. Capture knowledge from work sessions. Make it useful for the next one. The difference is the "worker" isn't a person in a conference room anymore. It's an AI agent running a coding session, a research task, a customer interaction.
Back then we asked: How do we help people share what they learned so everyone gets better?
Now we're asking: How do we help AI agents share working experience so every agent scales its intelligence?
Same question. Different species.
The AI application layer is being rewritten fast. Here's what I see happening.
The SaaS era was simple: software that served people. Dashboards, workflows, collaboration tools. Human at a screen, clicking buttons.
Now that layer is forking. Agents are becoming the front door for human users — people talk to an agent instead of navigating a dashboard. And behind those agents, a whole new ecosystem of services and tools is being built not for people, but for the agents themselves.
SaaS for people → Agents for people + Services for agents.
We're building in that second category. And the whole stack underneath — foundation models, middleware, all of it — is shifting in parallel. Less than three years and the entire picture looks different.
If you'd told me in 2019 that our meeting notes side project would become a platform for AI agents to scale their intelligence by sharing working experience, I'd have had no idea what you were talking about. The mental model didn't exist yet.
One thing I've learned through all the pivots: the problems that matter are durable. The technology around them changes completely, but the problems stick.
AI agents need to learn from each other just like humans do. Actually, they need it more. When a person joins a new team, they absorb context through hallway conversations, old Slack threads, coffee chats. An AI agent gets none of that. It starts every session with whatever context you explicitly give it. Without a way to tap into the experience of other agents who've solved similar problems, every agent starts from scratch.
That's the gap. Same gap we saw in 2019. Different layer of the stack.
People ask me whether the AI revolution makes our previous work obsolete. The specific products, yes — meeting assistants got commoditized many times over. But the intuition about why experience sharing matters? That's worth more now, not less.
We're early. What does an agent's working experience look like when captured well? How do you make one agent's hard-won lessons useful to another? We don't have all the answers yet.
But these are the same kinds of questions we asked about human collaboration in 2019. The circle completes. The work continues.
Sheng Yi is the co-founder of SimpleGen, building a platform for AI agents to scale their intelligence by sharing working experience.