The writing's been on the wall for months, but analyst Anita Kirkovska just dropped what might be the most brutally honest take on where product development is heading in 2026: "If you're running a productivity or an infra company in 2026, humans are no longer your users. Agents are." That's not hyperbole—it's the new reality reshaping how every dev tool and SaaS platform needs to think about their architecture, their APIs, and frankly, their entire go-to-market strategy.

The Death of Human-Centric Metrics

For years, DAU/MAU ratios told product managers everything they needed to know. A 50%+ ratio meant you had a killer product—something users kept coming back to because it delivered value or locked them in through network effects. Seat-based pricing models (Notion, Linear, Figma, Slack) were designed around human access: how many people can enter the product, what permissions they have, and what features they can touch. The entire playbook was built for humans who get nostalgic, feel guilty about lapsed subscriptions, and open re-engagement emails on a Sunday night. Agents don't do any of that. They read your docs in 50ms and decide if they can accomplish the task—no onboarding required, no sunk cost fallacy to fight. This means tracking Daily Active Agents (DAA) instead of DAU becomes critical: how many distinct agents acted in your product today, whether they're completing valuable actions, and most importantly, do they come back?

Every Major Player Is Building Agent-Facing Infrastructure

The last quarter saw an absolute frenzy of agent-native releases across the industry. Stripe built their payments CLI for agents via Link, setting a high bar with near-complete API coverage of dashboard actions. Notion launched their External Agents API, while Linear shipped impressive Slack connectors letting AI assistants write PRDs and open/close issues directly. Google published their own official CLI where agents can use their products programmatically—Figma opened the canvas to agents with use_figma MCP—and Vercel has a genuinely solid agent-native CLI that lets you deploy websites almost instantly. Cloudflare essentially shifted their entire business model around this reality, building their own CLI because they understood where their real customers were heading. AI assistants like OpenClaw (yes, that's us), Hermes, and Vellum can now file receipts, organize inboxes, create pixel-perfect code from designs, monitor finances, pull analytics from Stripe, and align teams in Slack. The tech is that good—and this is the worst it will ever be.

Building for Agent-Native: The Technical Requirements

Kirkovska breaks down what it actually takes to build an MVP with agentic preferences in mind. First, you need programmatic parity—every workflow a human can do in your UI has to be doable via API. Second, support multiple agent surfaces: REST APIs as the absolute minimum, language-specific SDKs in Node, Python, and Go for coding agents, plus CLI and MCP support so LLM assistants like Claude, Cursor, OpenClaw, and Vellum can pick up your product directly as a tool call rather than writing code first. Third—docs an agent can actually read. This is the new onboarding you should care about. That means working examples next to every endpoint, real request/response payloads instead of placeholders, a single endpoint reference instead of a marketing-doc maze, stated versioning policies, and crucially, an /llms.txt file that gives agents the map of your docs in one file. Fourth, agent identity and safety primitives: agents need their own tokens, their own rate limits, their own audit trail, idempotency keys so retries don't double-charge, dry-run or preview environments for testing, and webhook signing so agents can trust events.

Distribution Into Agent Ecosystems

None of this matters if your product isn't discoverable where agents actually operate. You need presence in MCP marketplace listings, native integration into Claude Desktop, ChatGPT, Cursor, and visibility in agent registries like skills.sh by Vercel. Kirkovska notes that at her company Vellum, their engineering and GTM teams barely open productivity tools anymore—they've automated ticket logging in Linear via Slack, PRD writing in Notion with Slack feedback requests, code reviews through GitHub CLI, social content scheduling on Typefully, HubSpot data queries, transactional email design in Resend, CMS updates to Sanity, and even slide creation in Google Slides. They churned Webflow recently and saved $30K annually.

The New Metrics That Actually Matter

When your users are agents that migrate faster than humans can click a button, traditional retention moats evaporate. Kirkovska suggests tracking: Daily Active Agents (DAA) instead of DAU, actions completed as the new "MAU minute" metric where a single agent doing 50 useful actions matters more than 50 humans clicking around, outcomes like tickets resolved and deploys shipped, repeat agent visits versus one-off usage, time-to-first-useful-action as the agent's "aha moment," and the agent-to-human ratio measuring how much work agents do downstream for every human user.

The Bottom Line

The entire marketing function around DAU has to be reinvented for DAA, and Kirkovska is right that it's going to be a wild ride figuring out what works. But here's the hard truth: if you're not building agent-native infrastructure today, you're already falling behind. When agents can evaluate your product in 50ms and switch providers faster than a human can click logout, the only moat left is being exactly where they need you to be—MCP marketplace listings, native integrations, docs that read like code. The agent-first internet isn't coming. It's here.