Eight years building Neptune’s developer GTM

This is not a consulting engagement. It is the operator story behind the consulting.

I spent eight years helping build Neptune’s developer go-to-market—from an early-stage product in a category that barely existed, through Series A and multiple shifts in the market, to the company’s acquisition by OpenAI.

During that journey, my role evolved from co-founder and DevRel to Head of Growth and eventually CMO.

I spoke about those things on Scaling DevTools

We entered a category that barely existed.

When Neptune started, experiment tracking and model metadata were not established software categories.

The people experiencing the problem were described as data scientists, machine-learning engineers, researchers, and eventually AI researchers. Those roles—and the problems most important to them—changed as the market developed.

Early marketing therefore had two jobs.

We had to explain the problem and establish why it deserved a dedicated product. At the same time, we had to avoid getting so abstract that the people doing the work could no longer recognize themselves in the message.

That tension remained throughout Neptune’s life: communicate a large enough vision for where the market was going while staying grounded in the product and problem we could solve today.

We chose a game we could realistically win.

Neptune had a strong competitor in Weights & Biases. They had deep Silicon Valley proximity, strong relationships with researchers, and a playbook built around that position.

We were a Poland-rooted company with a different set of cards. Trying to copy their conference and community strategy would have meant playing the same game from a weaker position.

So we looked for the opening they were not pursuing as aggressively.

We went heavily into technical content and search. We built deeply researched tutorials, comparisons, tooling landscapes, and practical articles around the problems machine-learning practitioners were already searching for.

The strategy was not generic thought leadership. It was to become genuinely useful across the MLOps landscape, build authority in adjacent categories, and introduce Neptune when the product was relevant.

The MLOps blog eventually reached 300,000 monthly unique visitors.

The company changed, so the positioning had to change.

Neptune’s market did not stand still.

At different stages, we focused more heavily on production machine learning, research workflows, and eventually teams training foundation models. Each shift required decisions about the product, audience, positioning, and website.

This was not a sequence of superficial rebrands. The question was always the same:

Who has the clearest and most expensive version of the problem right now—and what can Neptune credibly solve for them today?

The answer determined which use cases we emphasized, which product capabilities led the story, and which language belonged on the homepage.

It also taught me one of the principles I now bring into client work: your investor story and your customer-facing message cannot do the same job. Customers need to understand the concrete value available now, not decode the five-year vision first.

The website and product journey had to work as one system.

Traffic alone was never the final goal.

Once a developer discovered Neptune, the website had to make the product understandable. The documentation had to help them evaluate it. The onboarding experience had to get them to value without unnecessary friction.

We worked across the homepage and self-serve journey, treating messaging, product presentation, onboarding, and activation as connected parts of the same system.

That work helped increase Neptune’s self-serve activation rate by 3×.

The lesson was simple: discoverability creates an opportunity, but the rest of the journey determines whether that opportunity becomes meaningful adoption.

I also learned what I would do differently.

Eight years produced plenty of wins, but it also made the missed opportunities clearer.

Once we found that technical content and search were working, we probably diversified too early. We could have doubled down harder and continued until the channel was genuinely saturated.

I would also have taken more creative risks. In marketing, silence is often a bigger failure than an experiment that receives mixed reactions. Safe work can produce acceptable results, but the outcomes that meaningfully change a company tend to be uneven and disproportionately large.

Those lessons matter as much as the headline results. Operator experience is not knowing a perfect playbook. It is recognizing which decisions worked, which did not, and what you would change the next time.

From early stage to acquisition by OpenAI.

In 2025, OpenAI announced its acquisition of Neptune to strengthen the tooling and infrastructure used to understand frontier-model training.

The acquisition does not mean that every positioning decision, campaign, or experiment was right. And marketing alone did not create that outcome.

What it does mark is the end of a complete operating journey: helping a technical product find its market, building a repeatable distribution engine, adapting the message as the market changed, improving the self-serve journey, and supporting the company through to exit.

That is the perspective I now bring to developer-tool companies: not an outside theory of what growth should look like, but the judgment built by working through the entire journey from the inside.