What’s Working in DevTools Marketing Right Now? with Karl Hughes from Draft.dev
Karl Hughes surveyed DevTools marketers on budgets, AI, and ROI. The results: only 7% say AI is unlocking real value in marketing so far, influencer marketing has the worst measured ROI of any channel but the highest experimentation rate, and in-person events still win.
Karl Hughes runs Draft.dev, a content agency for developer tools companies. This year they ran their first formal survey of DevTools marketers — about 50 responses, partly collected in person at conferences, partly sent to clients and prospects — asking where budgets are going, what's actually working, and how marketing teams are really using AI day to day.
We get into the results: a sharp swing from constrained 2025 budgets to most teams increasing spend in 2026, why only a small fraction of marketers say AI is unlocking real value despite near-universal experimentation, how Draft.dev rebuilt its own production process around AI (cutting a four-to-six-week process down to two or three hours), and why distribution — not content production — is becoming the real edge.
The Q&A below is experimental — extracted from the episode transcript with AI. If you spot a mistake, let us know at jack@scalingdevtools.com.
What was this survey, and how did you run it?
We're always trying to get a pulse on where the DevTools marketing space is going — where people are spending, where they're getting ROI, how they're actually using AI. We ask this constantly on sales calls, and I probably do 100 to 200 of those a year, so I get a fair number of data points there. But this year we wanted something more structured and formal — our first time doing it. It was partly collected in person at conferences, partly sent out to clients and prospects. We got around 50 responses. Not a massive number, but realistically there are only a few hundred companies that are big enough in dev tools for this to matter, so I feel good about the coverage, even if it's not scientific-level.
What were the big budget takeaways, and was there a pattern in who was increasing spend versus cutting it?
2025 budgets were very constrained, especially in the first half — a lot of fear around tariffs and the broader economy, and because the US carries such outsized weight in the dev tools economy in terms of spending, talent, and funding, that uncertainty rippled through the whole industry. I talked to a lot of agency owners in this space, and almost everybody had that same pullback last year. This year is different — about 62% of the teams we talked to are increasing their budgets, most of the rest are holding steady, and only about 10% said they're likely to decrease, which is a very small number. My read is that new investment flowing into AI, plus a for-now more stable economy, has given people more confidence to spend on marketing. As for a pattern in who's increasing versus cutting — not really, and we ran the survey anonymously on purpose, so we're not trying to dissect specific companies. It was across the board: smaller startups with a good AI story and fresh funding tend to be spending more, and bigger companies that have started leveraging AI to ship better products are also feeling good about marketing spend.
Does AI growth actually hurt demand for dev tools — are more people just building their own alternative instead of buying?
Not that we're seeing, at least not broadly. In platform engineering there's some concern that companies will just build their own version of something like Backstage or Port rather than buy it, since for small-to-midsize companies those are fairly straightforward tasks to replicate — for big enterprise, that's probably less true. But overall, because there's money trickling into AI tools, those companies still have to buy developer tools behind the scenes. Even the big AI labs have hundreds or thousands of engineers building real software, deploying it, and running security checks on it — AI agents aren't replacing that work yet. So at the moment, the boom seems to be trickling down and benefiting developer tooling broadly. A lot of companies are also just reshaping their product story around AI — some of that's a fresh coat of paint, some of it's genuinely using AI in interesting ways.
How much value are marketing teams actually getting from AI right now?
There's a lot of experimentation, but no consensus on how it's best used and no standardization across companies. 96% of people said they'd played with AI — sure, we're all in tech, we hear about it every day. But only about 7% said it's unlocking significant value — saving lots of hours steadily, without a ton of manual overhead and rework. That surprised me; my own number for Draft.dev would be much higher. The biggest area where it's helping is written content, editing and planning — something like 60% of people use it for that consistently. But it still requires an actual subject-matter expert to catch hallucinations and make sure things are subtly correct, because we're not selling body soap, we're selling complex technical products that change quickly, and a lot of the source material LLMs are trained on is already a couple of years out of date. Marketers struggle to trust the tools because right now they're just not accurate enough, and that has to improve before we see this used at real scale in dev marketing.
Why is adoption inside marketing teams so uneven, given the tools exist?
A lot of companies haven't figured out how to actually deploy AI and let their teams use it productively. I talked to a marketer at a big cloud company who said certain tools are whitelisted, but there's no company account that lets her share things with teammates — so she's on Gemini, the person next to her is on ChatGPT, someone else likes Claude, and everyone's building their own siloed workflow with nothing standardized or sanctioned. At small companies like ours, there's basically no adoption curve — you try a tool, it saves time, your boss says yes. At bigger, more locked-down companies, people are often stuck on an older, worse version of a single sanctioned tool they're not allowed to upgrade, and the output is noticeably worse for it. It's easy to forget that even when the tools are genuinely good enough, adoption cycles inside real companies still take time — procurement and habits don't change just because the technology did.
You reworked Draft.dev's whole production model around AI. How did that change your output?
It used to take four to six weeks and several rounds of revisions with real human editors to get a single piece of content out. Now it's two or three hours with one engineer and a couple of LLMs in the loop, to the same quality, consistently. At a high level: we feed in context about the product, the competitive landscape, and a whole other stream of stylistic preferences and best practices — including things that make AI sound less like AI. A real engineer builds the initial prompt for a specific piece, a second model checks the first model's output, and the engineer does a final quality check on something that's already 75 to 80% of the way there — catching hallucinated facts, wrong version numbers, and making sure any code actually runs. Client comments also get fed back into our inputs for the next piece, so every revision cycle makes the next one better. It works on almost everything we produce, from hands-on tutorials to higher-level thought leadership — it struggles most when a client has literally just released something with no documentation or use cases indexed anywhere yet. Because of all this, our prices roughly halved last year — there was no reason to keep them where they were once we got that much more productive.
If content production keeps getting cheaper for everyone, where does the actual edge in marketing come from now?
Distribution. Putting out good technical content and showing up consistently when buyers ask an LLM what the best tool for X is — that's a great strategy right now, but it may not last another year or two once agents start actually doing the purchasing on people's behalf, and then we may go through another big shift in how you distribute for agentic workflows and payment cycles. One theory I have: human distribution — an actual following — is going to become a much more important asset over the next one to five years. If people start to realize LLM answers can be shaped by whoever produces the most content with the best distribution, they're going to fall back on their friends and the people they follow instead. That's part of why influencer marketing is getting so much experimentation right now: in our survey it had the lowest measurable ROI of any channel, but the highest experimentation rate, because it's a new channel without a known ceiling yet, and the payoff could be big. The single highest-ROI channel in the whole survey, though, was human connection — events and community meetups. As digital channels get more saturated and flooded, people go back toward in-person.
Tactically, how do you actually optimize content so it shows up in LLM answers?
You just need to think in terms of what a potential buyer of your product would actually ask ChatGPT. It sounds almost too simple, but that's pretty much it — the tools for getting real insight into the best prompts to optimize for are still limited, but directionally, if you build content that answers those actual buyer questions and distribute it across multiple sites, you tend to see lift. For distribution, we publish to a client's own site, syndicate similar versions to places like Dev.to, Medium, and individual blogs, get it shared across company and founder or employee LinkedIn pages, and do paid promotion through newsletters, which has become a more interesting channel than it used to be. Honestly, nobody's fully cracked this yet.
A generic "what is a data warehouse"-style explainer used to be a solid SEO play. Does that still work?
That kind of high-volume, low-competition, top-of-funnel content used to be a real strategy — we wrote a lot of "what is a data warehouse" style pieces over the years. The problem was never that the content was bad, it's that there's no real intent to buy your product just because you defined a term well. Google's shift toward AI-written zero-click summaries for exactly those high-level queries hit that strategy hard last year — a lot of clients who'd built traffic on those terms saw big drops. So now we're pushing more content toward the middle and bottom of the funnel, where buyers actually show up ready to act. My own buying behavior has shifted the same way — shopping for a microphone recently, I didn't Google "best microphone," I described exactly what I needed to ChatGPT and bought based on what it recommended. That's a much shorter path than reading a great explainer two years before eventually buying, so you have to optimize for that kind of specific, closer-to-purchase question instead.