Notion Is Un-Inventing Product-Led Growth. Its Newest Engineers Report to Sales.

Notion is hiring a Forward Deployed Engineer. The job sits on a team the company calls Services, and the person who gets it will spend most of their time inside other companies' offices.

The listing asks for someone who can write production code in front of a customer, stand up custom agents through MCP and Notion's Agent APIs, build data pipelines, run large content migrations, and argue with a client's security team about permissions. Prior professional services experience is a nice-to-have. Pay is quoted as total on-target earnings of $175,000 to $240,000, incentive pay included. The title ends in GTM.

Hold that next to Notion's own biography. This is the company that reached a reported 100 million users on word of mouth and community templates, with a free tier generous enough to work as the marketing budget. It is the example founders reach for when a board asks why they still have no sales team. Ivan Zhao and Simon Last rebuilt the entire product in Kyoto in 2015 on money borrowed from Zhao's mother, and what they came back with was software so obvious that selling it felt beside the point.

Ten years later, Notion is hiring consultants and paying them partly on commission.

Nobody at the company has described it that way, and there is no reason they would. But a job board is where a strategy change becomes visible first, months or years ahead of any financial statement, and this one is visible now.

A year of building toward it

Notion has been moving this direction since last September, when Notion 3.0 rebuilt the AI layer around agents instead of a writing assistant. Custom Agents went into beta in February. In May the company shipped a developer platform with an External Agents API and Workers for custom logic, and said a million agents had already been built on it. Zhao described the goal as any data, any tool, any agent.

The numbers gave him room to be aggressive. Notion's ARR was around $600 million by the end of last year, and executives have said AI accounts for roughly half of it, with paid AI attach climbing from about 20% to over 50% in twelve months. A tender offer in January, with Singapore's GIC participating, valued the company near $11 billion. COO Akshay Kothari has said the sales team is doubling.

That last one deserves a second look. Doubling a sales force is a routine decision at most software companies and an identity crisis at this one.

Why the free tier stopped working

Product-led growth ran on a property that most software happened to have. It explained itself. You opened a doc, typed a sentence, understood the point within a minute, and sent a link to three coworkers. That is what turned a free tier into an acquisition channel instead of a cost center.

Agents do not behave that way. An agent's output depends almost entirely on context the product cannot see, starting with your permissions model and the taxonomy your company actually uses, which is rarely the one written in the handbook. It also needs some working agreement about when a task counts as finished, and most companies have never written that down. Everest Group made the same point in July, arguing that an agent has to be fitted to a particular company's data and compliance situation before it returns anything worth having, that no two companies share that situation, and that this alone breaks the ship-once, configure-lightly economics SaaS distribution was built on.

Nobody works out their own taxonomy during a free trial. Someone has to come and do it. At Notion, that someone now sits in the go-to-market org.

The two costs SaaS spent twenty years removing

Software earned 75% to 85% gross margins because the second copy cost nothing to make and the second customer cost nothing to serve. AI has put a price back on both.

ICONIQ's 2026 research puts gross margins for AI products near 52%, up from 41% in 2024 and still well short of traditional software, and finds inference taking a larger share of spend as products mature. That is the compute side. On the delivery side, any deployment that needs a person to configure it becomes a cost that scales with the customer count instead of the code.

Notion already runs a meter for the first one. Custom Agents consume credits, and admins watch the consumption on a dashboard in real time. The Services team is the answer to the second. Neither line item existed in the business model that got the company to $11 billion.

Pricing follows. GitHub Copilot moved to usage-based billing in June, which we called the first honest price the product had ever quoted. The workspace layer is next.

The model left the labs

In June we covered AWS putting $1 billion behind a forward-deployed engineering organization, following joint ventures from Anthropic and OpenAI built on the same reasoning. An analyst summed up the mood at the time: the AI companies had looked in the mirror and decided they wanted to be Palantir.

Notion sells documents. As of May, one job-board analysis counted 118 companies hiring forward-deployed engineers, and the roster had already stopped looking exotic: Intercom, GitLab, Snowflake, Stripe, Brex, Cloudflare, Notion. Most of those are application software businesses that spent a decade finding ways to avoid employing people like this.

For anyone buying software right now, the useful question at the end of a demo has changed. Ask who the implementation engineer reports to, and whether any part of their pay moves when your account grows.

The skeptic's case

Two arguments say this is a phase.

Inference keeps getting cheaper, faster than almost anything else in computing. GPT-4-class capability ran roughly $20 per million tokens in late 2022 and about $0.40 by early 2026. A company living with 52% margins today may get most of that back by 2028 without touching its price list.

The second argument sits inside the job description. Notion wants its forward-deployed engineers to leave behind migration tooling, integration patterns, reference implementations and internal delivery tooling. The company is hiring people to build the thing that makes those people unnecessary. On that reading, Services is scaffolding around a temporary gap rather than a permanent department. Constellation Research's Ray Wang has raised a related worry about "fake FDEs," engineers who drift into expensive sales support instead of doing the technical work the model depends on.

Both arguments hold up over a decade. The next three years are the problem, and they are the stretch in which a generation of bottom-up software companies has to learn a motion it was founded to avoid, on a cost base it was valued for never having.

What Notion is betting

There is a version of this that ends well. Palantir spent ten years arguing that embedded delivery creates switching costs a subscription never will, and it has the returns to support the argument. An engineer who built your permissions model and half your agent library is harder to tear out than a document editor. On that path Notion is trading margin for grip, deliberately.

The other outcome has the company walking toward an IPO window carrying a services attach rate, a token bill and a commission plan for engineers, then asking public markets to keep valuing it like software. Airtable came out of the same bottom-up tradition and sold to Bending Spoons this year for less than it raised.

Which way it goes depends on things Notion does not control. One question is already closed. The product stopped selling itself, and the company said so on its own careers page.