A two-person startup called Poke just raised $4 million to pursue one of the most deceptively simple ideas in artificial intelligence: what if building an AI agent were as easy as sending a text message?
That’s the pitch. No code. No drag-and-drop workflow builders. No complex prompt engineering. Just a group chat interface where humans and AI agents collaborate in plain language, passing tasks back and forth the way coworkers do on iMessage or WhatsApp.
The seed round was led by Craft Ventures, with participation from Y Combinator, General Catalyst, Soma Capital, and a roster of angel investors that includes former GitHub CEO Nat Friedman and Perplexity AI CEO Aravind Srinivas, according to TechCrunch. For a pre-revenue company with two employees, that’s a notable lineup of backers — and a signal that serious people in Silicon Valley believe the current generation of AI agent tools is still far too complicated for the average knowledge worker.
Poke’s co-founders are Andrew Milich and Ananay Arora, both former engineers at companies where they watched firsthand how non-technical teams struggled to adopt AI tooling. Their thesis is blunt: the interface is the bottleneck. Most agent-building platforms today assume users are comfortable with programming concepts — variables, conditional logic, API integrations. Even the so-called “no-code” solutions require users to think like software engineers, constructing visual flowcharts that mirror the underlying logic of code without actually writing it.
Poke throws all of that out.
Instead, users create what the company calls “threads” — essentially group chats populated by one or more AI agents alongside human participants. A user might type, “Research the top five competitors in the European EV charging market and put the results in a spreadsheet,” and an agent handles it. Need a second opinion? Add another agent to the thread with different instructions or a different model powering it. Need a human sign-off before the agent sends an email? The human stays in the loop, literally, inside the same conversation.
“We wanted to make the interaction model something everyone already understands,” Milich told TechCrunch. “You don’t need to learn a new tool. You already know how to text.”
It’s a clever framing. And it arrives at a moment when the AI agent market is simultaneously exploding with funding and stalling on adoption. Venture capital has poured billions into companies promising autonomous AI workers — from Cognition’s Devin (valued at $2 billion) to Adept AI to scores of smaller startups building agents for sales, customer support, legal research, and finance. But actual deployment inside enterprises remains thin. A recurring complaint from CIOs and IT leaders: these tools are impressive in demos and frustrating in practice.
The gap between what agents can theoretically do and what non-technical employees can actually get them to do is wide. Poke is betting that gap is primarily a design problem, not a capability problem.
There’s historical precedent for this kind of bet paying off. Slack didn’t invent workplace messaging. It made it feel intuitive and even enjoyable, and that interface advantage was worth $27.7 billion when Salesforce acquired it. Notion didn’t invent wikis or project management databases. It wrapped them in a flexible, attractive interface that made people want to organize their work. The underlying technology mattered, of course. But the interface determined adoption.
Poke’s group-chat metaphor carries a specific structural advantage: it makes multi-agent orchestration legible to humans. One of the hardest problems in the current wave of AI agent development isn’t getting a single agent to perform a task — it’s coordinating multiple agents on complex, multi-step workflows while keeping humans informed and in control. Most platforms handle this through backend orchestration layers that are invisible to end users, which creates a black-box problem. Users don’t know what the agents are doing, why they made certain decisions, or where things went wrong when they inevitably do.
Poke’s chat threads surface all of this by default. Every action an agent takes appears as a message in the conversation. Every handoff between agents is visible. Every point where a human needs to intervene is obvious because it looks like what it is — someone being asked a question in a group chat. The transparency isn’t a feature bolted on after the fact. It’s a byproduct of the interface itself.
That said, skeptics will raise legitimate questions about how far a chat-based interface can scale. Simple tasks — research this, draft that, summarize these documents — map naturally onto conversational interactions. But what about workflows with dozens of conditional branches, error-handling routines, and integrations with enterprise systems like Salesforce, SAP, or Workday? At some point, the argument goes, you need the precision and structure that visual builders or code provide. A chat thread can become just as chaotic and hard to follow as any Slack channel with too many participants.
Milich and Arora appear aware of this tension. According to their comments in TechCrunch’s report, Poke is initially targeting small-to-medium businesses and individual knowledge workers — not Fortune 500 enterprise deployments. The play is bottom-up adoption: get individual users hooked, let them build agents for their own workflows, and let the product spread organically within organizations. It’s the same go-to-market strategy that worked for Dropbox, Figma, and, yes, Slack.
The competitive field, though, is crowded and getting more so by the week. OpenAI has been aggressively expanding its own agent capabilities, with its Operator product and the recently enhanced GPT-based assistant tools. Google’s Gemini agents, Anthropic’s Claude tool-use features, and Microsoft’s Copilot Studio all offer varying degrees of agent-building functionality aimed at non-developers. Then there are the dedicated startups: Relevance AI, CrewAI, AutoGen, LangChain-based agent frameworks, and dozens more. Many of these are also pursuing “no-code” or “low-code” positioning.
What separates Poke, at least on paper, is the purity of its commitment to the messaging metaphor. Other no-code agent builders still look like software tools. Poke looks like a chat app. Whether that’s a feature or a limitation will depend on execution.
The $4 million raise is modest by current AI funding standards — barely a rounding error compared to the multi-billion-dollar rounds going to foundation model companies. But seed rounds aren’t about scale. They’re about proving a thesis. Craft Ventures, which led the round, has a track record of early bets on developer and productivity tools; partner David Sacks has been vocal about his conviction that AI agents represent the next major software distribution wave.
And the angel investor list speaks volumes. Nat Friedman, who ran GitHub during the launch of Copilot, understands better than most how AI tools get adopted by technical and non-technical users. Aravind Srinivas has built Perplexity into one of the most-used AI products by obsessing over interface simplicity. Their participation suggests Poke’s approach resonates with people who’ve actually shipped AI products at scale.
The timing matters, too. Enterprise interest in AI agents surged through late 2025 and into 2026, but so did disillusionment with overpromised autonomous capabilities. A February 2026 survey by Gartner found that while 78% of enterprise technology leaders planned to deploy AI agents within 12 months, only 14% had successfully moved a pilot project into production. The most commonly cited barriers were not technical limitations of the AI models themselves but rather integration complexity, lack of user trust, and difficulty training non-technical employees to use agent-building tools effectively.
Poke is positioning itself as the answer to that last barrier. If agents are going to become ubiquitous — embedded in every business process from procurement to customer onboarding to internal HR requests — the tools for creating and managing them need to meet users where they are. And where most users are, every single day, is inside a messaging interface.
There’s a philosophical dimension here worth noting. The dominant metaphor for AI agents over the past two years has been the employee — an autonomous digital worker who receives instructions, executes tasks, and reports back. This metaphor implies hierarchy and delegation. Poke’s group-chat model implies something different: collaboration. The AI agent isn’t below you in an org chart. It’s beside you in a conversation. That’s a subtle shift, but it changes how users relate to the technology. Delegation requires trust and clear instructions upfront. Conversation allows for iteration, correction, and back-and-forth refinement in real time.
Whether this distinction matters to buyers and users outside of San Francisco remains to be seen. Most business software purchases are driven by concrete ROI calculations, not interface philosophy. Poke will ultimately need to demonstrate that its chat-based agents can reliably complete tasks that save people real time and money — and that the simplicity of the interface doesn’t come at the cost of capability.
For now, the product is in early access. The company hasn’t disclosed user numbers or revenue figures, which is unsurprising for a two-person team that just closed its seed round. The next twelve months will be telling. Can Poke attract enough early users to build momentum? Can it handle the inevitable feature requests that push against the constraints of a chat-only interface? Can it build integrations with the enterprise tools that business users actually rely on?
Big questions for a small company. But sometimes the smallest ideas — make it work like texting — turn out to be the biggest ones.
WebProNews is an iEntry Publication