👋 Hi, it’s Rohit Malhotra and welcome to the FREE edition of Partner Growth Newsletter, my weekly newsletter doing deep dives into the fastest-growing startups and S1 briefs. Subscribe to join readers who get Partner Growth delivered to their inbox every Wednesday morning.

If you’re new, not yet a subscriber, or just plain missed it, here are some of our recent editions.

Small teams don't have time to stitch together five tools and hope it works.

Apollo gives you everything you need to find leads, reach them, and close deals — all in one place:

Stop juggling tools and start building pipeline that scales.

With Apollo, the AI revenue engine powering 4M+ users.

Sign up for free

_{Interested in sponsoring these emails? See our partnership options here.}

Subscribe to the Life Self Mastery podcast, which guides you on getting funding and allowing your business to grow rocketship. 

Previous guests include Guy Kawasaki, Brad Feld, James Clear, Nick Huber, Shu Nyatta and 350+ incredible guests.

In March 2026, a texting app called Poke launched.

Four months later, Cognition bought it for nine figures.

Poke never turned a profit. It lost money on every message. And it just sold to the startup behind Devin, the AI software engineer.

Here is what most people will miss.

Cognition is not buying a chatbot with a sense of humor. It is buying the one thing its team of competitive programmers was never going to build: an agent people actually want to talk to.

Devin can write code. It cannot make you laugh. Poke can.

100 million messages in three months. Hundreds of thousands of users. A product that felt less like software and more like a friend.

The model was never the moat. The interaction layer is.

And Cognition put a nine-figure price on that idea, exactly one year after it bought Windsurf.

Every AI company is trying to build a better model.

Poke built something harder to copy. A personality.

It does not wait for commands like a tool. It texts you like a friend. Slang. Jokes. It messages first.

And it lives where you already are: iMessage, SMS, Telegram, WhatsApp. No app to download. In June 2026, it became the first AI agent Apple approved for its Messages for Business platform.

In three months, hundreds of thousands of people ran travel, health, finance, scheduling, and education through it. 100 million messages. The most common use was the boring stuff that sticks: email, reminders, to-dos.

Now look at who bought it.

Cognition was started by three programmers who won gold at the International Olympiad in Informatics. They began in crypto, pivoted to AI, and built Devin, an autonomous software engineer.

Devin is very good at writing code. It is also very much a robot.

Here is the part money cannot fix. Making an agent feel like a colleague is not a compute problem. It is a product problem. And Poke had already solved it, while losing money on every message.

So Cognition did not license the model. It bought the team.

Nine figures. For a company four months old.

That is what Cognition agreed to pay for The Interaction Company of California, the maker of Poke. Co-founder Marvin von Hagen confirmed the price sits in the low nine figures. Everything below that stayed private.

There is no revenue multiple to quote here. Poke never disclosed revenue, and by its own account struggled to make any. So price it against the only thing Cognition bought before this.

One year ago, Cognition bought Windsurf. That deal came with $82 million in ARR and more than 350 enterprise customers. Real revenue. Real accounts.

Poke came with none of that. 100 million messages and a personality.

Same acquirer. One year apart. Two completely different assets.

And nothing changes for Poke through the end of 2026. It stays live. It stays on Apple's platform. The real integration is pushed to next year, when Poke starts running on Cognition's newest model, SWE-1.7.

Von Hagen and his team join Cognition and keep running the product.

Read that last part slowly. The team that could not afford the compute just got the compute. And kept the keys.

Three numbers do the explaining here. None of them is revenue.

One. Nine figures for a company that launched in March 2026. Four months of life. No disclosed profit. Set it against Windsurf, undisclosed but estimated near $250 million, which arrived with $82 million in ARR and 350-plus enterprise accounts. Poke arrived with none of that. The math says Cognition did not buy a business. It bought an interaction model and the team that authored it.

Two. 100 million messages in three months, across hundreds of thousands of users. That is engagement, not income. And engagement is the asset. Cognition already has Devin ARR that ran from $1 million in September 2024 to $73 million in June 2025. What it did not have was a product people text like a friend. The 100 million messages are the only proof that justifies the price.

Three. $10.2 billion. That is the valuation Cognition raised its $400 million round at in September 2025, led by Founders Fund. Nine figures for Poke is a line item against that mark. The math worked before the ink dried, same as it did with Windsurf a year earlier.

Poke does not get folded into Devin on day one. Nothing changes through the end of 2026. It stays live, stays on Apple's platform, and von Hagen and his team join Cognition and keep running it. The real integration is pushed to next year, starting with Poke adopting SWE-1.7 for some tasks.

That is the stated plan.

Cognition used almost the same language once before. When it bought Windsurf in July 2025, the message was that the team would continue to operate as they had been. Within three weeks, Cognition laid off 30 people and offered buyouts to the roughly 200 who remained. That was a remnant team left after an acquihire, not a founder-led product bought whole, so the parallel is not clean. But the pattern is worth naming. At Cognition, continue as you are has a shelf life measured in weeks, not quarters.

What Cognition is not disclosing is the internal build this replaces. No public roadmap, no named team, no shipped feature that gives Devin memory across sessions or lets it run more than one task at a time.

Poke built a product hundreds of thousands of people loved without ever having the infrastructure to run it profitably. It lost money on every message, competing for consumer attention against the same players who own the phone by default. That ceiling is real for any consumer AI app without a model and a balance sheet behind it. Von Hagen also gets to keep running it. He and his team join Cognition and keep serving Poke's users, distribution and survival without losing authorship of the product.

Wu's team gets what no internal Devin roadmap could deliver before its rivals closed the gap on model quality. An interaction model people already text like a friend, plus the team that built it, slotting into a coding agent that is powerful and lifeless. It comes with a specific unlock von Hagen has already named: Poke orchestrating multiple Devin sessions and carrying memory across them, turning one agent into a coordinator of many.

There is one more detail that explains why this closed clean. Wu and co-founder Walden Yan were angel investors in the company before they bought it. This was not a process won on price. It was a relationship that started with a check.

Cognition is wagering on four things being true at once.

One. The interaction layer is becoming as valuable as the model. How an agent talks, stays proactive, and remembers you is where differentiation moves once frontier models converge, and they are converging. Cognition is betting that owning how the agent behaves, not just how it reasons, is what wins the next round.

Two. Buying the team that nailed it beats building it. A company of competitive programmers building an autonomous software engineer is not the natural home of consumer-grade, messaging-native personality. Cognition is betting it is faster to acquire that muscle than to grow it, the same call it made with Windsurf a year ago.

Three. Poke breaks Devin's ceiling. Devin runs one pull request at a time. Cognition is betting Poke becomes the layer that runs parallel Devin sessions and carries memory between them, solving Devin's biggest limitation by acquisition instead of a multi-quarter build.

Four. The relationship de-risked the price. Wu and Yan were angels in the company. Cognition is betting that buying a product it knew from the inside, run by a team it had already backed, beats a cheaper deal for strangers.

Standalone consumer AI assistants lose first. Poke was the breakout of the text-your-AI-like-a-friend category. First agent approved on Apple's Messages for Business, hundreds of thousands of users, 100 million messages in a quarter. And it still could not stand alone, because the compute cost more than the product earned. The verdict on consumer-assistant-as-a-business just got written in an exit, not a shutdown, which is the more expensive way to learn that engagement without economics is a feature, not a company.

Cognition's coding-agent rivals lose second. Cursor pushed to a reported $500 million in ARR competing on model quality and IDE. Copilot competes on distribution. Cognition just bought a vector none of them are chasing, how the agent talks, remembers, and coordinates. When the models converge, the company that already owns the interaction layer sets the terms of the comparison.

Cognition's own internal team building a personality or orchestration layer loses quietly. Whatever roadmap existed to give Devin memory and a voice is now redundant, closed by acquisition rather than by whichever engineers were racing to ship it before this deal made that race unnecessary.

Three threads worth pulling.

One. The interaction layer just got a real-money price tag, not another slide about agents feeling human. Cognition's answer to a commoditizing model market was not an internal feature. It was a nine-figure acquisition of the team that already built the interface. The market gets to decide over the next year whether owning how an agent talks is a moat or a rounding error.

Two. Cognition is assembling a full-stack agent company acquisition by acquisition. Windsurf in July 2025 for the IDE, the enterprise ARR, and the talent. Poke in July 2026 for the interface, the engagement, and the orchestration. Exactly one year apart. Fin was not an isolated move for Salesforce, and neither is this for Cognition. Watch what gets bought next before this year ends.

Three. The exit window for consumer AI founders just changed shape. Poke launched in March 2026 and sold for nine figures roughly four months later. No revenue. No decade to IPO. An engagement-and-team story that resolved in a single quarter. The clock for a breakout consumer app is no longer years to profitability. It is months until the infrastructure company that needs your DNA decides it is cheaper to buy you than to build you.

The bear case is real. Founder-led teams absorbed into larger acquirers have a long history of attrition, and the playbook of a loved product getting buried under a bigger company's roadmap is older than Poke itself. The looseness that let Poke use slang and crack jokes is exactly what an infrastructure company's process tends to sand off. Cognition's own precedent is not gentle. When Windsurf got the continue-as-you-are language in July 2025, layoffs and buyouts followed within three weeks. Different situation, same gravity. And there is no clean multiple here to anchor the price, only nine figures against a product with no revenue.

The bull case is sharper. This was never about Poke the texting app. It is about the team's product sense and one specific unlock. If Poke becomes the conductor over parallel Devin sessions, with memory that persists across them, Cognition bought its way past Devin's single biggest limitation instead of engineering around it for quarters. The nine figures is not really the bet. The bet is that von Hagen's team makes Devin feel like a colleague faster than Cognition's own engineers, or its rivals, would have allowed.

What to watch by the end of 2027: one signal decides it. If Cognition ships Devin orchestration, parallel sessions and cross-session memory, built by the Poke team, the thesis converted. If Poke is still a standalone texting app with no Devin integration shipped and the founders have moved on, nine figures bought a talent park, not a moat. The tell is whether SWE-1.7 running under Poke turns into Poke running over Devin. The first is an integration. The second is the whole reason for the deal.

Here is my interview with Tal Kirschenbaum, co-founder and CEO of Ledge. He's not building software that tracks the month-end close. He's building AI agents that that tracks the month-end close.

If you enjoyed our analysis, we’d very much appreciate you sharing with a friend.

Here are the options I have for us to work together. If any of them are interesting to you - hit me up!

And that’s it from me. See you next week.

What do you think about my bi-weekly Newsletter? Love it | Okay-ish | Stop it