Finding the rhythm: first impressions of a new role in a new corner of AI
Two weeks into a new DevRel role and I’ve learned more about what I don’t know than what I do. That’s probably the point.
The surprise
I spent the last few years in speech-to-text. The developer community there talks about accuracy rates, language coverage, realtime latency. The problems are about getting audio in and text out, fast and clean.
The new role is on the text-to-speech side. Same voice industry, completely different conversation. The community here talks about voice quality, emotion control, pricing per character, streaming audio boundaries. I keep catching myself framing problems from the old angle and having to flip my perspective. It’s humbling in a good way.
What I was right about
Content cadence matters more than content volume. Publishing one solid piece that takes real time beats three rushed roundups. The posts that got the best response were the ones with working code, clear explanations, and something useful at the end. The changelog-style summaries got polite nods. The difference is measurable.
Also right: data wins arguments. A pricing comparison table with real numbers at the top of a public leaderboard generated more real conversation than any opinion piece I’ve written. Developers trust numbers they can verify.
What I’m still figuring out
How to balance the two audiences. There are people building on the platform I work for who need reference docs and migration guides. And there’s the broader AI community that wants industry analysis and comparisons. They overlap but they’re not the same group, and serving both without watering either one down is harder than I expected.
Also figuring out how fast to react to market moves. Google changed pricing mid-week. A Chinese lab shipped a new model variant. A competitor restructured their tiers. Each one is a story but reacting the same day every time is a treadmill. I’m experimenting with a weekly roundup that groups them into context rather than chasing each one individually.
The thing I keep coming back to
Voice AI quality has converged enough that the decision factors are moving to price, latency, languages, and ecosystem. That makes DevRel work different from two years ago, when part of the job was convincing people the tech was ready. Now the tech is ready and the job is helping people choose wisely and build well. I like that version of the job better.
Still plenty to figure out. That’s the fun part.