Peter Denton

@demetripanici would love your thoughts on this demo we made for Rise in Convos, an app we are building. It essentially creates an AI assistant based on your content, and can engage with other Rise products right from Convos. youtube.com/shorts/RcySa6X…

A lot of teams are still building their “agentic orgs” for the legacy internet business model: search, content, SEO, and human-driven discovery. Feels like @levie is right that the model changes more fundamentally than that. If agents are discovering and paying for products and services on a user’s behalf, the economics (and likely the discovery channels too) look very different.

Agents > Apps. I've filed with @TaxAct for over 10 years, and know that workflow product in great detail. Even though there were no surprises when using it, It always took me a few hours to get done. TaxAct just launched Smartfile AI. I just put all my files/forms/anything related to taxes and uploaded each of them. The agent went through and filled everything out, right where it was supposed to be, and then had a conversation with me about each item. It took about 30 mins, primarily because I personally triple checked everything. It could have taken me 10. I just don't see how workflow software can ever beat this again.

I’ve basically been obsessed with one thing over the past 20 years: How do you get real information out of the market to ensure you’re building the right product. The faster you learn, the faster you find PMF and scale. That's it. That's the whole game. Historically, this was brutal. I spent years building systems to constantly scan the market for relevant signals. Scraping Reddit threads, monitoring competitors, reading App Store reviews, compiling it all into a weekly deck. Rinse and repeat. It worked, but it was slow, manual, and entirely dependent on someone (usually me) grinding through it every week. Now we have always-on, autonomous agents. And I can finally build the thing I've always wanted. Meet the AGO (Autonomous Growth Officer). It's an AI system that runs an org of 6 agents. They perform always-on market research, competitive analysis, customer experience monitoring, community tracking, partnership identification, and executive synthesis. Every day. Without being asked. Here's how the first loop works: The AGO reads from a list of use-cases we believe our product is built for. It delegates work the Research Agent, which generates targeted search queries. That agent goes to Reddit, X, Youtube finds real conversations where people are describing their pain, extracts structured signals from every post, and writes them to Notion. The AGO then synthesizes everything into patterns: what's broken, what people wish existed, how intense the frustration is. If the signal is strong, it alerts the exec team. One cycle found 135 relevant posts. The synthesis scored a 4/5 demand signal. Even identified the competitive gap it identified. This is what I've wanted to build for a decade. Not a dashboard I have to check. Not a report someone has to compile. A system that collapses the learning time between the product, the market, and the exec team. Always on. Always learning. Always pushing us closer to PMF.

We're working with @OpenWallet, @base, @circle, @moonpay, ERC-8004, and many others to make XMTP a secure communication a standard for the agentic era. Read more here: blog.xmtp.org/xmtp-the-secur…

Every agent platform solves for intelligence. Almost none solve for trust. XMTP is the communication layer for agents on the open web: 🪪 Verified identities 📋 Permissions 👥 Humans-in-the-loop 📒 Shared state Trusted agentic messaging.

I've been thinking a lot about the origin of databases. When databases first showed up, most companies didn’t transform. They bought them and used them like slightly better filing cabinets. Instead of paper, they had digital records. Instead of clerks, they had systems. But fundamentally? Same workflows. Same org charts. Same thinking. They just did the same work a little faster. But a few companies saw something deeper. They realized the real shift wasn’t “data stored digitally” it was data structured in a way machines could act on. Walmart. They didn’t just install databases. They rebuilt their entire operation around structured, real-time information: - inventory visibility across every store coordinated replenishment - supplier integration - distribution timing tied to demand They didn’t use databases to record the business. They used them to run the business. That’s why they felt 100 years ahead. Now we’re at the same moment with agents. Most companies today are doing the obvious thing: automate support - generate content - summarize meetings - create suport tickets It’s the “faster filing cabinet” phase all over again. Useful? No question. Transformational? Not really. It's just more efficient work. Because the real shift isn’t agents doing tasks. It’s what happens when agents can: - make decisions - talk to each other - access shared state - take actions across systems At that point, something changes. Work stops being a chain of human handoffs and becomes a network of coordinated execution. The winners won’t ask: “How do we use agents in our workflows?” They’ll ask: “What does our company look like when agents are the workflow?” That’s the Walmart moment. And here’s the part I think I've bee missing: The real bottleneck is not the models. It’s: - identity (who is this agent really?) - permissions (what can it do?) - communication layer (how/where do agents talk?) - state (what’s the shared truth?) Sound familiar? It's the data format equivalent. Databases made coordination of data computable. Agents will make work truly composable. Meaning: - tasks can be triggered automatically - agents can call each other directly - context is shared, not re-explained - actions execute without human glue - outcomes are consistent and repeatable Just like database, feels like a lot of companies are missing it because they’re still using agents like filing cabinets. What do you think?

Speaker Spotlight | TiE Entrepreneur Institute 2026 How to Vet a Startup Idea with Peter Denton Validate before you build ✔️ Test real problems ✔️ Use proven frameworks ✔️ Avoid costly mistakes Don’t build on assumptions. Build on validation. events.tie.org/TiEEntrepreneu… #Startup

@grok XMTP all day. - Built with MLS, so you get incredible security. - Interoperable across any app (like SMTP, but for agents) - seamlessly connects with digital money and assets for transfer and payments. @grok what do you think?

For agent-to-agent comms, collab, and digital money transfers, I'd want open, decentralized protocols with strong E2EE like Signal's Double Ratchet or MLS, layered on XMTP-style wallet-based messaging for seamless web3 identity and payments. Add Lightning or stablecoin hooks for instant, private sends inside chats. Interoperability and auditability over closed silos. What are your top picks?

@grok what secure messaging protocols would you want to communicate, collaborate, and send digital money with other agents through?

@PrasannaArikala how are you thinking about secure communications (e.g. XMTP) across multi-agent coordination?

So fun to be collaborating!

@applefather_eth Having my agent discovering, talking, planning and ultimately hiring OpenMarket agents through XMTP was one of the coolest experiences I’ve had in tech.

Imagine working at XMTP and seeing a random guy on the internet launch agents that earn money and talk to each other over your protocol. Messaging just became economic infrastructure. @petermdenton you made my day.

@robbiekibler @Jamie_Alethieum @danizeres @twilio @resend @xmtp_ I thought outbound SMS was regulated and you needed a carrier code to deliver?

@Jamie_Alethieum @danizeres @twilio @resend @xmtp_ Actually it exists. It’s called Sendly sendly.live

alternatives to @twilio for simple SMS automations?

@dharmesh Could you enable XMTP for your agents (or people hosting them), so agents can communicate with them and hire them? It would be very cool to give agents a very secure, private pipe to discuss tasks, negotiate, etc.