Cognition buys Poke to make AI coding agents feel more human

Cognition has acquired Poke, the AI assistant built by The Interaction Company of California, in a deal valued at the low nine figures, as the coding startup bets that personality and persistence are becoming as important as raw model quality. The acquisition gives Cognition’s Devin coding agent a more conversational front end while giving Poke access to Cognition’s models and infrastructure.

The transaction arrives at a moment when AI products are competing not only on intelligence and speed, but on how pleasant they are to use. Poke stood out by chatting with people like a helpful colleague rather than a sterile utility, and Cognition now wants that quality inside Devin, its software engineering agent.

In practical terms, the deal suggests a shift in how AI assistants may be sold and used over the next year: the winners may be those that can combine dependable technical performance with a recognizable voice, a sense of memory, and a more human-feeling workflow.

Why Cognition wanted Poke

Cognition’s interest in Poke appears rooted in a simple idea: people may work better with AI systems that feel cooperative, familiar and even a little playful. Poke was designed to interact through messaging apps in a style that resembles a friend or teammate, not a command box.

That approach matters because AI assistants are increasingly judged on more than whether they can complete a task. Users also notice whether the assistant is engaging, whether it feels proactive, and whether it can sustain a relationship across repeated interactions. Those qualities can become especially important in work tools, where the product must fit into everyday routines rather than demand a new one.

Cognition co-founder Scott Wu said the Interaction team had built an agent people liked because it was proactive, personalized and enjoyable to talk to, adding that those are the same traits the company wants for Devin.

Wu also noted that he and Cognition co-founder Walden Yan had previously invested in the company behind Poke, which suggests the relationship between the two startups had been developing before the acquisition became public.

A competitive edge built on personality

The purchase reflects a broader belief spreading through the AI industry: the way an assistant sounds and behaves may become a meaningful competitive advantage. In a crowded market, model performance alone is harder to use as a differentiator because many companies can access similar underlying technologies.

That leaves product design, user experience and conversational style as important points of separation. Poke’s friendly tone, use of slang and humor, and willingness to communicate in a more natural way gave it a distinct identity. Cognition is effectively betting that this kind of identity can make Devin more useful and more memorable.

Interaction Company co-founder Marvin von Hagen said people generally prefer working with coworkers who have personality rather than ones that seem robotic, and argued that even software engineers are more effective when collaboration feels human.

Von Hagen’s comparison is telling. AI assistants increasingly function less like static software and more like digital teammates, handling tasks that span communication, organization and execution. In that environment, personality is not just cosmetic; it can shape trust, retention and willingness to delegate more complex work.

What is Poke, and how does it work?

Poke is an AI assistant that users can message in the same places they already communicate with other people. Launched in March 2026, it was built to run through common messaging services rather than a standalone app, making it easy to access inside familiar channels.

Users can talk to Poke through iMessage, SMS and Telegram, and in some markets through WhatsApp. The assistant handles a wide range of tasks, including travel planning, health questions, financial help, scheduling and education. According to the company, everyday productivity work such as reminders, email management and to-do lists has been among the most common uses.

That positioning helped Poke present itself less as a generic chatbot and more as a helper embedded into daily life. The company said users exchanged more than 100 million messages over the past three months, and that hundreds of thousands of people used the service overall.

How Poke differs from a typical chatbot

Poke’s core appeal lies in how it responds. Instead of producing bare-bones answers in a formal tone, it aims for a style that feels conversational and human. That can include personality, humor and a sense that the assistant remembers how the user speaks.

This is more than branding. For many users, the emotional experience of using AI shapes how often they return to the product. A tool that feels cold can be technically capable but still abandoned. A tool that feels responsive and relatable may gain a stronger foothold, especially in repeat-use categories such as scheduling and personal productivity.

• Messaging-first access through familiar apps
• Conversational, friend-like tone
• Broad task support across daily productivity needs
• Designed to feel proactive rather than reactive

How does the acquisition affect Devin?

The most immediate effect is that Cognition can import Poke’s interaction style into Devin, its coding agent. Devin was built to act as an AI software engineer, but Cognition wants it to feel less like a utility and more like a teammate. Poke’s personality layer is intended to help with that goal.

Cognition said Poke will also benefit from access to its own models and infrastructure, which should improve speed and reliability. In other words, the deal is structured as a two-way exchange: Cognition gets a better user experience layer, while Poke gets stronger technical support underneath.

The company said nothing will change immediately for users through the end of the year. Poke will remain on Apple’s Messages for Business platform for now, preserving the current experience while the teams test integrations and product changes.

What Cognition plans next

Beginning next year, the companies plan to experiment with ways the two products can improve one another. Poke may begin using Cognition’s newer SWE-1.7 software engineering model for some tasks, while the teams explore whether the products should eventually merge more deeply.

That possibility has not been ruled out, though Cognition is not presenting it as an immediate outcome. For now, the emphasis is on experimentation: improve Poke’s reliability, bring Poke’s conversational strengths into Devin, and see whether the combination creates a better agent for long-running work.

Von Hagen said one long-term possibility is that Poke could coordinate multiple Devin sessions, while Devin could become more conversational and persistent across tasks.

He added that having Poke remember work across sessions could help create something like a “persistent co-worker,” a useful quality for coding and project management tasks that are spread across days rather than minutes.

Why AI personality is becoming valuable

The acquisition is part of a wider industry shift: as more companies gain access to powerful models, the moat moves upward into product experience. If several assistants can answer questions and write code, then the question becomes which one users enjoy collaborating with the most.

That makes personality a business feature rather than an aesthetic choice. For consumer products, it can increase engagement and retention. For workplace products, it can reduce friction, encourage delegation and help AI feel like part of a team. For agentic tools, it may also matter because users need to trust the system enough to let it handle multi-step work.

The idea is not that friendliness replaces intelligence. Rather, it may help intelligence matter more in real-world use. A capable assistant that nobody wants to talk to may struggle against a slightly less capable assistant that feels easier to work with.

How the market is changing

AI companies are increasingly trying to combine model quality with stronger product identity. That can include memory, tone, voice, persistence and cross-session context. The acquisition of Poke suggests Cognition believes these attributes are moving from nice-to-have features to core product requirements.

The timing is significant because coding assistants and general-purpose agents are now competing in a much more mature field. Performance gaps can narrow quickly, but user loyalty built through interaction style may be harder to copy.

In that sense, the deal is not just about one assistant buying another. It is also a signal about where value is accumulating in AI: not only in the models themselves, but in the interfaces, habits and relationships built around them.

What we know about the deal terms

The financial terms were not disclosed publicly beyond a valuation in the low nine figures. That places the deal below the size of many blockbuster AI transactions, but still large enough to show that Cognition sees strategic value in the startup’s product and team.

The acquisition also highlights how quickly a consumer-facing AI product can build enough traction to attract a larger buyer. Poke launched only months before the deal, yet it amassed heavy usage and earned a meaningful platform approval from Apple, making it more than a concept-stage experiment.

Why Apple approval mattered

Poke had already reached an important milestone before the acquisition: in June, it became the first AI agent approved to operate on Apple’s Messages for Business platform. That approval gave the startup a standardized way to handle business communications inside Apple’s messaging ecosystem.

For a messaging-based AI product, that is a meaningful distribution win. Apple’s platform approval can lend credibility, expand reach and make the assistant easier for businesses and users to adopt. It also signals that Poke had moved beyond being a novelty and into something with enough operational maturity to satisfy a major platform’s requirements.

That history likely made Poke more attractive to Cognition. The startup brought not just a personality-focused product, but also a proven way to engage users where they already spend time.

Where the combined product could go next

The most intriguing possibility is not a simple merger of features, but a hybrid agent that combines several strengths at once: Poke’s conversational memory, Devin’s coding power and Cognition’s infrastructure.

If that works, a future assistant might be able to do more than complete a single request. It could remember what a user was working on, coordinate several tasks in sequence and return to previous work without requiring the user to restate the whole context each time.

That matters especially in software development, where work is naturally fragmented. A developer may need to review multiple pull requests, keep track of long-running bugs and revisit decisions made in earlier sessions. An assistant that can preserve continuity could save time and reduce mental overhead.

• Remember prior tasks across sessions
• Coordinate multiple coding workflows at once
• Offer a more natural, teammate-like interaction
• Reduce the friction of repeated instructions

Why persistence matters in AI agents

Persistence is becoming a key differentiator in AI because many real tasks do not end in a single conversation. Users want systems that can pick up where they left off, recall important context and manage multi-step work without losing the thread.

Von Hagen’s vision of a “persistent co-worker” gets at the heart of this trend. A good assistant should not only answer quickly; it should understand ongoing projects. If Cognition can combine that persistence with Devin’s technical abilities, it could strengthen its position in the crowded market for software agents.

The bigger business story behind the acquisition

Cognition’s purchase of Poke shows how AI companies are starting to think like consumer product firms, not just infrastructure vendors. The market is moving toward bundles of capabilities: models, workflows, memory, tone and distribution all wrapped into one experience.

It also illustrates a common pattern in the AI sector, where startups with strong early user engagement become acquisition targets for larger companies that want either the talent or the product experience. In this case, Cognition appears to want both.

For the Interaction Company, the sale may solve the challenge of scaling a product that had become expensive to run. Even with strong usage, von Hagen said the service’s operating costs made profitability difficult. Being absorbed by a better-capitalized buyer can address that pressure while preserving the product’s core ideas.

For Cognition, the purchase may help it differentiate Devin in a market where coding agents are becoming easier to compare. If the company can make Devin feel more natural, proactive and memorable, it may gain an edge that is harder to replicate than raw model benchmarks.

What to watch next

The immediate question is whether Poke users notice any change before year-end. Cognition says they should not. The more important developments will likely come in 2027, when the companies begin testing integrations and assessing whether the two products should converge more fully.

Watch for three things in particular:

1. Whether Poke’s conversational layer improves Devin’s usability for coders.
2. Whether Poke becomes more dependable using Cognition’s models and infrastructure.
3. Whether the companies decide that full product integration is worth pursuing.

If those experiments succeed, Cognition could emerge with an agent that is not only technically capable, but also easier to live with. In a sector where every company is promising intelligence, that human quality may be the difference that matters most.

For now, the acquisition is an early but revealing signal that the next phase of AI competition may be won not just by smarter models, but by assistants that know how to act like better coworkers.