Eka Robotics Raised $150M at a $1.45B Valuation for a Robot That Feels. The Vision-Force-Action Model Is the Architecture That Finally Makes Dexterous Manipulation Scale.

On October 19, 2026, Town became an AI unicorn after one of the fastest ARR ramps in enterprise software history. The company raised a $108 million Series B at a $1 billion valuation, led by Forerunner Ventures and Menlo Ventures, after reporting that it hit seven figures in annual recurring revenue within weeks of its product launch. The round brings total disclosed funding to approximately $160 million — $55 million in a June Series A, $108 million now — for a company that was essentially pre-revenue when it last raised capital.

The product that drove that growth velocity is deceptively simple to describe: every employee gets their own AI assistant, called a Townie, connected to their email, documents, calendar, and the business systems they personally work in. The Townie knows the employee's current projects, open commitments, past decisions, and working relationships. It does not ask the employee to summarize their situation before answering a question. It already knows the situation.

This description makes Town sound like a better ChatGPT plugin. The new coordination feature announced with the Series B reveals what it actually is: a network of per-employee AI agents that can query each other, hand off work between themselves, and share context across teams — without the human employees having to relay any of it. That is not a feature increment. It is a different architecture for how AI works inside an enterprise.

The Problem with Horizontal AI Tools

To understand what Town is building, it helps to understand what every enterprise AI tool built before 2026 is not building.

ChatGPT Work, Microsoft Copilot, Google Gemini Enterprise: these are horizontal tools. Every user gets the same generic AI capability applied to their query at the moment they ask it. The model knows nothing about who the user is, what they worked on last week, or what their current commitments are. Every session starts from scratch. The user's productivity gain is bounded by how much context they can efficiently summarize in a prompt.

Signal's analysis of Microsoft Copilot's enterprise activation gap found that 93.5% of Microsoft 365 users eligible for Copilot had not converted to paid AI seats. One of the most consistent reasons cited in enterprise buyer interviews was context reset: the AI did not know the user's work, so the user had to explain their situation every time, which negated much of the productivity benefit for tasks requiring organizational context. You can ask Copilot to draft an email to Sarah, but you have to explain who Sarah is, what your relationship is, and what the last exchange covered. A human assistant who has worked with you for three months does not need that explanation.

Town's thesis is that the context layer is the product. The Townie is not a better model — it uses models from multiple AI labs, the same models available to Copilot and ChatGPT Work. The Townie is a better-contextualized agent, connected to the specific data sources that define what a particular employee does, making decisions and drafting responses from an accurate model of that employee's work state.

What a Townie Actually Does

Each Townie connects to the employee's work ecosystem through a permissioned integration layer: email, calendar, documents, Slack or Teams history, CRM records, project management tools. Unlike a generic AI tool that accesses these through one-time API calls at query time, the Townie maintains a persistent contextual model of the employee — a running summary of current projects, active commitments, key relationships, and recent decisions, updated continuously as new data arrives.

When an employee asks "What do I need to follow up on from last week's call with Acme?", the Townie does not search email and return raw results. It knows that the Acme call happened, who was on it, what the open action items are, whether any have been completed, and which are approaching deadline. It answers with a prioritized list of the employee's specific open items, not a list of documents about Acme.

This capability compounds over time. Town states that nothing a customer connects is used to train models — which means the Townie's context model is a private representation built and maintained by Town's infrastructure on the customer's behalf, not a contribution to a shared training dataset. The longer a Townie has been connected to an employee's work streams, the more accurate its context model, and the more it can do without explicit instruction.

Notion's experience with 24-hour agents established that extending agent runtime from minutes to hours dramatically changes what autonomous AI can accomplish — but only when the agent has access to accurate context. A 24-hour agent with no knowledge of the employee's current work state is a powerful tool in search of a task. A Townie with accurate context is an agent that can autonomously work the queue while the employee is in a meeting.

The Team Mesh: Agents Coordinating with Agents

The Series B announcement's most significant feature is the team mesh: a layer of AI-to-AI coordination that allows Townies to query each other directly.

In practice: a salesperson's Townie can ask the engineering team's Townie whether the feature the customer is asking about is on the roadmap and when it ships. The engineering Townie looks at current sprint data, checks the roadmap, and responds with a specific answer — without the salesperson having to send a Slack message, wait for it to be seen, and wait for a reply. The entire information exchange happens in the background.

This changes the architecture of enterprise knowledge work at the AI layer. Instead of AI tools that augment individual employees independently, Town is building a network where AI agents share context and coordinate tasks across the organizational structure. A shared business update — a new product announcement, a customer escalation, a policy change — automatically propagates through the team mesh, updating each Townie's context model without requiring anyone to forward an email or update a shared document.

The business impact is concentrated in cross-functional workflows: the places where information handoffs between teams create the most friction and the most delay. Legal review of a contract, engineering input on a sales proposal, customer success context for a renewal conversation — all currently require human-to-human communication to relay organizational context. The team mesh makes that context exchange happen at the AI layer.

Why This Funding Happened So Fast

Town raised a $55 million Series A in June 2026 and a $108 million Series B at a $1 billion valuation in October — four months later. That pace is unusual even in a high-velocity AI funding environment. The explanation is in the ARR trajectory: seven figures within weeks of launch is a growth velocity that VCs see perhaps once or twice per year.

RevenueCat's 2026 data on AI app churn showed that AI tools have a 79% annual churn rate — driven primarily by context reset and lack of persistent value delivery. Town's per-employee model addresses that churn driver directly: a Townie that accumulates context over months becomes harder to cancel precisely because replacing it means losing the context it has built. The switching cost compounds over time, which is the kind of retention mechanic that enterprise investors find compelling.

Forerunner Ventures and Menlo Ventures co-led the B alongside returning investors a16z and First Round Capital. Forerunner's focus on consumer-inflected enterprise products and Menlo's enterprise software depth represent complementary validation of both the market positioning and the go-to-market mechanics.

The Per-Employee Model vs. the Horizontal Tool

The enterprise AI market in 2026 has sorted into three distinct architectures, and understanding the differences matters for procurement decisions:

The per-employee architecture has the highest activation cost — it requires connecting multiple data sources for each employee — and the highest retention value, because the context moat compounds. Horizontal tools have the lowest activation cost and the lowest retention because every session resets. Role-specific tools sit in the middle: strong context within their domain, limited capability outside it.

Town is betting that knowledge workers' most valuable AI use cases are the cross-functional ones — the questions that require context from multiple systems and multiple relationships — and that horizontal tools can never address those cases well because they have no persistent context. The Series B valuation prices in the possibility that this bet is correct.

The Competitive Landscape

Town is competing against three categories simultaneously:

Horizontal AI tools (ChatGPT Work, Gemini Enterprise, Claude.ai Enterprise): larger, better-funded, distributed through relationships enterprises already have. Their advantage is model capability and ecosystem integration. Their disadvantage is context reset — the problem Town directly solves.

AI productivity platforms (Notion AI, Confluence AI, Microsoft Copilot): strong document context within their product domain and zero context outside it. A Notion AI knows about documents in Notion; it knows nothing about the employee's email, external meetings, or CRM data unless specifically integrated.

Enterprise AI automation platforms (Relevance AI, Make, n8n): automate workflows but do not build a persistent agent with employee-specific context. They are tools, not coworkers.

Town's moat is the combination of breadth (connecting all work systems, not just one) and persistence (building context over time, not resetting per session). Neither characteristic is technically difficult to replicate — any company could build this. But the context data is the moat once it accumulates, and the team mesh coordination layer adds a network effect: the more employees on the platform, the more useful the coordination layer becomes.

Claude Duet's launch — Anthropic's proactive agent that acts without being asked — represents a different answer to the same activation problem Town is solving. Claude Duet inverts the interaction model; Town builds a persistent context layer under the existing interaction model. These are not directly competitive approaches, but they are competing for the same enterprise AI budget line: the always-on AI assistant category that is replacing the AI tool I use when I remember to category.

The Activation Playbook for Per-Employee AI

For enterprise product and IT teams evaluating per-employee AI platforms, the deployment experience differs significantly from horizontal AI tools.

1. Start with a single team, not the whole organization. Per-employee AI requires data integration setup for each employee. Deploying to 10 people in a single function — a sales team, a customer success team — allows the integration pattern to be debugged before scaling. The context model needs 2–3 weeks of data accumulation before the assistant starts being meaningfully useful.

2. Connect the highest-signal data source first. Email and calendar are typically the highest-signal data sources for context because they reflect the employee's actual commitments and relationships. Connect these first, before CRM or project management tools, and measure query quality improvement after two weeks versus the baseline from day one.

3. Define three high-value use cases before deployment. The per-employee model is flexible by design, but employees without a specific mental model of what to ask will default to using it as a generic search tool — which is not its differentiating capability. Define three use cases specific to the team's work: summarize open action items from the past week; draft a follow-up based on this meeting; check what the engineering team says about this feature request. These anchor the team's mental model of what the Townie is for.

4. Measure context depth separately from query volume. The standard AI tool metric — queries per user per week — is the wrong metric for a per-employee context platform. The right metric is context depth: how many of the user's active projects does the Townie have accurate information about? Track this separately and use it as the leading indicator of whether the platform is delivering differentiated value.

5. Plan the team mesh rollout as a second phase. The coordination capability requires Townies from multiple employees to be sufficiently context-rich to give useful answers to each other's queries. Do not activate team mesh coordination until individual Townies have accumulated at least three weeks of context data. A team mesh connecting shallow Townies produces generic answers; a team mesh connecting rich Townies produces genuinely useful cross-functional coordination.

6. Treat Townie context as an offboarding artifact. When an employee leaves, their Townie contains a rich context model of current projects, open commitments, and key relationships. Define a policy for what happens to that context at offboarding — transferred to the incoming employee, archived for the manager, or deleted — before deployment, not after the first departure.

What Town's Unicorn Valuation Signals About Enterprise AI

Town reaching $1 billion in enterprise value in approximately one year of operation is a data point about where enterprise AI product-market fit is emerging. The horizontal tools have massive distribution and massive funding, but their churn rates tell the story: RevenueCat's 2026 data showed 79% annual churn for AI applications, driven primarily by generic tools that do not deliver compounding value. Enterprise buyers are discovering that a generic AI tool everyone has access to but few use regularly is not a productivity transformation — it is a procurement decision that looks good in a board presentation and delivers modest ROI.

Per-employee AI addresses this at the structural level. The context moat makes the tool more useful over time rather than less. The team mesh creates coordination value that only exists when the platform has broad enterprise adoption — a network effect that supports premium pricing and low churn simultaneously.

The $1 billion valuation at Town's ARR level implies a multiple justifiable only if investors believe the retention economics will prove dramatically better than horizontal tools. That is exactly what the product architecture is designed to produce. The test will be whether per-employee AI's context-building mechanism is durable enough — whether the Townie stays useful as the employee's work changes, projects close, and new ones open — or whether the context model degrades without continuous re-grounding to current reality.

For enterprise procurement teams, the right question is not whether to buy a per-employee AI platform but what is the true cost of context reset in current AI tool usage, and how much of that cost would a persistent context layer eliminate. The answer tells you exactly what Town's product is worth to your organization.

Takeaway

Takeaway: Town's $108 million Series B at a $1 billion valuation is the strongest market signal yet that per-employee AI — a persistent, personalized agent connected to each employee's actual work context — is becoming the enterprise standard rather than horizontal tools that reset context with every session. The platform's seven-figure ARR within weeks of launch, combined with the team mesh coordination layer enabling agent-to-agent knowledge exchange, validates the thesis that persistent context is the missing layer in enterprise AI. The deployment playbook requires more integration work upfront than generic AI tools, and the right activation metrics are different — context depth rather than query volume — but the retention economics of a context moat that compounds over time address the 79% annual churn rate that horizontal AI tools produce. For enterprise product and IT teams, the procurement question is now which AI architecture matches the organization's use case: horizontal tools for commodity tasks where context reset is acceptable, or per-employee AI for the cross-functional, context-heavy work where persistent knowledge of the employee's situation determines whether the AI is genuinely useful or just a fancy search box.