explainx / blog
Cognition, maker of Devin, acquired Interaction (Poke) on July 23, 2026. Here's why a coding-agent company bought a texting assistant — and what it means for both products.
Jul 9, 2026
SWE-1.7 from Cognition scores 42.3% on FrontierCode 1.1 Main — within points of GPT-5.5 and Opus 4.8 at fraction of cost. Kimi K2.7 base, 1000 tok/s in Devin, RL pipeline that challenges the post-training ceiling narrative.
May 22, 2026
Google I/O 2026 marks a paradigm shift in Search. From 24/7 background agents to search-driven coding and personal intelligence across 200 countries, discover the future of AI Search.
Jul 23, 2026
Robert C. "Uncle Bob" Martin — Clean Code author, coding since the late 1960s — says he no longer reads the code his AI agents write. Instead he constrains the agents and reviews a layered test pipeline. Here's exactly what that pipeline looks like, and where it might not generalize.
Cognition — the company behind Devin, the autonomous software engineering agent — announced on July 23, 2026 that it has acquired The Interaction Company of California, makers of Poke, an AI agent that lives inside your text messages. The announcement, posted at cognition.com/blog/interaction, is short on financial detail and long on strategic framing: Devin and Poke, Cognition says, are two faces of the same idea.
The obvious reader question is: why would a company that just spent a year proving Devin can migrate a bank's core codebase turn around and buy a consumer texting bot? The second question follows immediately — does this mean Cognition is drifting away from the enterprise coding business that made it worth billions? The short answer to both: no pivot. It's a bet that the same underlying architecture — an agent that works continuously, without waiting to be prompted, and reports back to you — is the right foundation for an engineer's coding assistant and a consumer's everyday life assistant.
| Question | Answer |
|---|---|
| What happened? | Cognition acquired Interaction (maker of Poke) — announced July 23, 2026 |
| What is Poke? | An AI agent that lives in iMessage/WhatsApp/SMS — handles flights, calendar, photos, smart home, health tracking |
| How big is Poke? | 100M+ messages exchanged in the last 3 months; hundreds of thousands of users; launched March 2026 |
| What's the Apple milestone? | First and only third-party AI agent approved on Apple's Messages Business Chat |
| How much did Cognition pay? | Undisclosed — no terms published in the announcement |
| Does this change Devin? | No — Cognition frames it as one "always-on cloud agents" thesis, not a pivot |
| Is my Poke access changing? | No — existing users keep the product unchanged; Cognition's infra should make it faster |
| Who's involved? | Scott Wu (Cognition CEO), Walden (Cognition co-founder, early Interaction investor), Marvin von Hagen (Interaction co-founder) |
Poke isn't a chatbot you open when you need something. It's designed to feel like a friend who texts you first — following up on a flight you mentioned, nudging you about a calendar conflict, or checking in on a fitness goal without being asked. Since its public launch in March 2026, Poke has exchanged more than 100 million messages with users in just the trailing three months, across hundreds of thousands of users, according to Cognition's announcement.
The number that matters more than the message count, though, is a platform milestone: Poke is the first and only third-party AI agent Apple has approved to text natively on Messages Business Chat, per reporting from AppleInsider and 9to5Mac dated June 4, 2026. That's a meaningfully higher bar than shipping an iOS app. Most AI agents — including most agent harnesses covered on explainx.ai — live behind a browser tab, a CLI, or a dedicated app icon a user has to remember exists. Poke lives inside the single thread on a phone that's already open dozens of times a day: iMessage. That's a distribution moat, not a feature moat, and distribution moats are much harder to copy.
Poke's task surface reads like a personal-assistant checklist: daily planning, calendar management, flight booking, photo editing, smart home control, health and fitness tracking — all mediated through text rather than a dashboard or app UI.
Cognition's framing for the deal is a single sentence worth unpacking: Devin and Poke unite under a shared vision of always-on cloud agents. What does that mean architecturally?
Contrast it with the typical chat-first assistant pattern: you open a window, type a prompt, wait for a response, and the agent goes dormant the moment you close the tab. An always-on cloud agent inverts that relationship. It runs continuously in the background — checking on tasks, reasoning over long horizons, and messaging the user rather than waiting to be messaged. Devin already works this way for code: it takes an async task, works for hours or days, and reports back with a pull request. Poke works the same way for daily life: it notices something needs attention and texts you about it.
Seen that way, the acquisition isn't "coding company buys chat app." It's "company that has spent two years building infrastructure for agents that work unsupervised in the background acquires the team that already proved that pattern works for 100 million consumer messages." Cognition's stated near-term plan reflects that: existing Poke users keep the product unchanged, while Cognition's models and infrastructure subsequently improve its speed and reliability. Poke's team, in turn, brings messaging-native UX patterns — proactive follow-up, conversational task completion — that could plausibly show up in how Devin communicates with engineers over time.
Skepticism about "always-on agents" is fair — plenty of agent products oversell autonomy. Cognition's answer on the coding side is its Nubank case study, published on devin.ai/customers/nubank. Nubank handed Devin its core ETL migration: an eight-year-old monolith spanning six million lines of code, roughly 100,000 data-class implementations, and a project that had involved more than 1,000 engineers historically.
The migration was originally projected to take 18 months. Cognition reports that verified business units — Data, Collections, and Risk — completed their migrations in weeks, not months, after engineers created migration examples and fine-tuned Devin as a specialized migration agent, with humans in the loop for approval at each step. Cognition's reported numbers: a 12x efficiency improvement in engineering hours and over 20x cost savings versus the original plan.
That case study matters for reading this acquisition correctly. Cognition isn't a coding-agent startup that ran out of enterprise traction and pivoted to consumer. Devin's enterprise business — the kind of workload covered in explainx.ai's top 10 agent harnesses guide and what is an agent harness breakdown — is scaling on its own. The Poke acquisition reads as additive, not a hedge.
It's also worth reading against Cognition's own recent model work: SWE-1.7, the frontier coding model Cognition shipped inside Devin on July 8, 2026, is evidence the coding side of the business is investing in frontier-model training, not standing still while leadership chases a consumer acquisition.
The people context here is unusually personal for an acquisition announcement. Scott Wu, Cognition's co-founder and CEO, wrote the blog post himself and disclosed that he and Cognition co-founder Walden were early investors in Interaction — meaning this deal has roots that predate the public announcement by a long stretch. Wu described the Interaction team as "exceptional people who move extremely fast and aim for only the most ambitious products."
Marvin von Hagen, an Interaction co-founder, replied on X that he "couldn't be more excited to start working together." Wu added a detail that frames the relationship as more than a transaction: von Hagen was reportedly on set for the original Devin launch video back in 2024, and Wu called the acquisition "two years in the making." Around the same time, Poke's own account posted that after almost two years in Palo Alto, Interaction is moving to San Francisco — "some opportunities are bigger than sunshine."
Taken together, this looks less like an opportunistic acqui-hire and more like two teams that have known each other since Devin's earliest days finally combining under one roof.
A few things worth holding onto before treating this as unambiguously good news:
Cognition's acquisition of Poke-maker Interaction isn't a pivot away from Devin's enterprise coding business — the Nubank case study shows that business is real and scaling. It's a bet that "always-on cloud agents," agents that work continuously and message you rather than waiting to be prompted, is a single underlying architecture that works for both an engineer's repository and a consumer's calendar. Poke's Apple Messages Business Chat approval gives Cognition a rare distribution channel that most AI agents don't have. The risk is organizational: two products with very different users, very different data sensitivity, and now one leadership team. Whether "always-on agents" is a coherent product thesis or a convenient narrative for a deal already two years in the making will show up in whether Poke gets measurably faster and more reliable — and whether Devin picks up any of Poke's proactive, messaging-native habits.
Sources: Cognition — cognition.com/blog/interaction · TechCrunch coverage of the acquisition · AppleInsider and 9to5Mac reporting on Poke's Messages Business Chat approval (June 4, 2026) · Devin customer case study at devin.ai/customers/nubank · posts on X from @interaction (Poke) and Scott Wu.
Reported as of July 23, 2026, based on Cognition's own announcement and contemporaneous press coverage — acquisition terms were not disclosed at publication time and may be updated as more information becomes available.
