DevTune is the GTM intelligence platform built for dev tool companies.

Track AI search visibility, LLM referral traffic, SDK adoption, and competitive positioning - all in one programmable, agent-first platform.

Please support our sponsors!

Every piece of software you've ever used assumed you were there.

A person at a keyboard, reading a screen, clicking things. Every UX principle, every onboarding flow, every pricing metric assumes a human is on the other end.

It's pretty clear now that this assumption is breaking.

The primary user of a lot of software, maybe most of it, is going to be an agent. Not a person. The sooner companies figure this out the more likely they are to succeed.

Think back to the early SaaS days. Software moved from on-premise installations managed by IT departments to browser-based tools used by individual employees. That shift, from IT buyer to end user, was the insight behind product-led growth. The user became the driver of adoption.

Now it's going the other direction. Tasks that needed a person at a keyboard are increasingly being handled by AI agents running on their own. Browsing a tool, pulling data out of it, kicking off a workflow. Agents are doing this stuff now.

An agent doesn't need a beautiful UI or onboarding tooltips. It doesn't care about your NPS score. What it needs is a reliable, machine-readable interface. Consistent outputs. A way to authenticate, act, handle errors, and know what it's allowed to do.

Agents need software built for them.

The design requirements are genuinely different.

Human-first software optimises for visual clarity, discoverability through UI navigation, onboarding that builds habits, re-engagement nudges, and packaging that appeals to a buyer's willingness to pay.

Agent-first software optimises for something else. Versioned APIs with predictable behaviour. Machine-readable docs. Clear permission models. Idempotent actions. Consumption-based pricing. Audit trails. Low latency.

Authentication is a good example of how different the requirements are. OAuth was designed for a human clicking "Allow" in a browser. Agents need something different - delegated auth with scoped permissions, time-limited tokens, audit trails showing what acted on whose behalf. This isn't solved yet, and it's one of the reasons "just add an API" isn't enough.

Most software today, even the good stuff, is built almost entirely for the first set. The second is barely an afterthought.

I think it happens in overlapping waves.

Agents are already becoming power users of existing software. They're hitting the Salesforce, HubSpot, Notion, Jira, and Slack APIs. Pulling data, writing data, completing tasks that used to need a person. The software wasn't designed for it, but agents are using it anyway, friction and all.

Next, existing software starts rebuilding for agents. Companies will notice a growing share of their API calls coming from agents, not humans. They'll add agent auth, better structured outputs, usage-based pricing for automated access, dedicated agent docs. When you're immersed in it day-to-day as I am it's easy to forget that we're at the very early edge of this. But it’s moving much faster than I thought it would.

And eventually, new companies get built agent-first from day one. This is where I think the really interesting stuff comes from. Instead of bolting agent support onto a human-first product, these companies build everything around the assumption that the primary consumer is an autonomous system. Their go-to-market won't be PLG in the traditional sense. I've been calling it agent-led distribution, for lack of a better term.

What will that look like?

Instead of optimising for a human discovering your product through search or word-of-mouth, you're optimising for an agent selecting your service from a registry because it has the best-documented API, the most reliable uptime, and a pricing model that works at machine scale.

Your marketing becomes your machine-readable metadata, and your onboarding becomes your schema definition.

This isn't theoretical. There are real signals right now.

The most obvious and visible is MCP (Model Context Protocol). MCP gives AI agents a standardised way to connect to external tools and services. It's plumbing for an agent-native ecosystem. We've seen this before: when REST APIs became standard, a generation of API-first companies got built on top of them. MCP feels like that kind of infrastructure moment, and adoption has been fast.

Then there's the infrastructure for agents to find and use other agents. Google's A2A protocol and cloud marketplace let agents discover and call each other directly. OpenAI and various startups are building agent platforms and registries. The pattern is familiar — think early app stores — but instead of humans browsing a catalogue, machines are negotiating with machines. Worth noting: MCP and A2A solve different problems. MCP connects agents to tools. A2A connects agents to each other. We're in a multi-protocol moment with no single standard for the full stack. If the app store analogy holds, we're somewhere between the Palm Pilot and the iPhone — the platform war hasn't been won yet.

Usage-based pricing is another signal worth watching. A single agent can consume more of your product in an hour than a person does in a month. Per-seat pricing just falls apart. Companies like Stripe, Twilio, and Snowflake are (maybe accidentally) well-positioned here, since they're already API-first and consumption-priced.

There's also a generation of headless, API-first products that were ahead of this without knowing it. Contentful, Algolia, Twilio. The interface was always the API and the UI was secondary. That made them natural fits for developer workflows, and now it makes them natural fits for agent workflows too.

On the dev side, agentic coding is already real. Claude Code, Codex, Pi, OpenCode, Forgen and the numerous other agentic coding harnesses can take an issue, write a plan, generate code across a repo, and open a PR. Every tool a developer uses is now potentially a tool an agent uses. That changes the requirements for every dev tool in the stack.

And from the incumbent side: Salesforce Agentforce. Salesforce is saying out loud that the future of CRM isn't a human logging into a dashboard, it's an agent acting on their behalf. When the world's largest CRM company makes that bet, you have to take the direction seriously, whatever you think of their execution (or their chances of pulling it off).

Here's what I'd be asking myself.

Who, or what, is going to use your product in three years? If you can picture a meaningful chunk of your product's functionality being run by an agent on a human's behalf, design for that now. Not later.

Is your API a real product, or a footnote? In an agent-native world, your API (and by extension your CLI) is your product. If it's underdocumented, inconsistent, or treated as secondary to the UI, you're already behind.

Does your pricing survive agent-scale usage? Per-seat pricing assumes a human in a seat. If agents are hitting your product at machine speed, what happens to your unit economics?

Ask yourself: are you building for agent discoverability? For humans, discoverability meant SEO and word-of-mouth. For agents, it means optimising AEO, being in the right registries, having machine-readable docs, and supporting standards like MCP or whatever replaces it. Your GTM probably needs a new layer.

And if you're wondering who's most exposed: companies whose primary value is a nice UI on top of an API call. If an agent can hit the same underlying service directly, the wrapper layer disappears. Per-seat SaaS with thin moats and no programmatic access is in a similar spot - the human in the seat was the only reason the business model worked.

I’m seeing growing aligned sentiment on this topic:

PLG was built on a specific insight: the end user, not the buyer, drives adoption. Serve them well and revenue follows.

The agent era inverts this. The user doesn't need to be served. They need to be enabled. And increasingly, the user is a machine.

To succeed in this new reality means taking this seriously across the whole business, from how the product works to how it gets distributed. In relative terms, almost nobody is doing this yet - it's a target rich environment. But it won’t stay that way for long.

Book a free 1:1 consultation call with me - I keep a handful of slots open each week for founders and product growth leaders to explore working together and get some free advice along the way. Book a call.

Sponsor this newsletter - Reach over 8000 founders, leaders and operators working in product and growth at some of the world’s best tech companies including Paypal, Adobe, Canva, Miro, Amplitude, Google, Meta, Tailscale, Twilio and Salesforce.

Check out Demand & Expand happening in SF in May (use Code PLGEEK to save 20%)

Check out our apps:
> Forgen - Multi-agent dev orchestration with human governance.
> DevTune - GTM intelligence for dev tools