The Changelog: Software Development, Open Source · Changelog Media

Selling SDKs in the era of many Claudes (Interview)

February 19, 2026·1 hr 50 min·8 clips

Steve Ruiz describes how Claude built an 80% complete starter kit in two hours, something he expected to take weeks.

1. The Changelog interviews Steve Ruiz, creator of TL Draw, a free whiteboard application and high-performance canvas SDK with approximately 130,000 registered users and half a million monthly visitors. 2. Steve Ruiz built TL Draw after contributing to Excalidraw and seeing the architectural limitations of its open-source codebase — he identified that an SDK with pluggable shape types would be more extensible, and built TL Draw around that insight. 3. The episode's central argument is that AI coding agents have eliminated time as the primary bottleneck in software development, but replaced it with harder, less tractable constraints around product positioning, community, and organizational alignment. 4. Steve describes finishing projects he had planned for Q3 2026 in the first week of January, and now having to reevaluate the company's strategic assumptions every six weeks as the parameters of what's possible keep shifting. 5. The conversation opens with a sponsor spot for Augment Code (Augie), which host Jared describes as one of his daily AI coding drivers alongside Claude Code — with Augment's 'context engine' cited as the differentiator. 6. An Augment Code executive appears in the sponsor segment to argue that the real story in the AI coding assistant market is not Cursor's fast revenue growth but Anthropic's position as the underlying model provider — framing Cursor as 'selling discounted tokens.' 7. Steve describes the psychological experience of running multiple AI coding agents as producing a specific form of guilt: lying awake concerned that he forgot to queue up tasks for his Claude agents before going to bed, and having 'confused partner' conversations about why this feels urgent. 8. Both Steve and host Jared frame this emotional state as a combination of fear and excitement inseparable from each other — likening it to a gambling addiction loop where the danger and the reward are the same stimulus. 9. Steve recounts prompting Claude Code to design and build a ComfyUI-style asynchronous image pipeline starter kit for TL Draw, forgetting about it, and returning two hours later to find it 80% complete — a project he had expected to take two focused engineers several weeks. 10. The five-day total shipping cycle for that starter kit, compared to Steve's original estimate of weeks, is presented as a concrete illustration of what roadmap compression means in practice for a small product team. 11. Steve describes TL Draw's 'zero backlog' exercise: having multiple AI tools triage every open issue and pull request to the point where the team can ask whether there's any reason not to close or prototype everything in the backlog. 12. TL Draw runs Claude Code on a Mac mini in Steve's living room every night to check the last 24 hours of merged pull requests and generate an up-to-date release notes document — a public commitment Steve says he would never have made without automation because manual maintenance would require dedicated product management resources. 13. Steve distinguishes between 'leaf node' work — starter kits, documentation, internal tools — where agents can run at full speed without dependencies — versus core SDK components with real user dependencies that require careful human oversight. 14. For internal tools, Steve describes using Gemini to synthesize raw CRM data from HubSpot meeting notes and deal logs into executive summaries, replacing what he calls the equivalent of having 'an army of consultants' available on demand. 15. Granola is mentioned as Steve's meeting notes tool, though he acknowledges he never reads the summaries directly — only when feeding them to another LLM to generate coaching insights or product feedback synthesis. 16. Steve frames technical founding as a particular advantage in the current moment because the technical founder simultaneously understands the problem being solved and the class of problem that AI tools can address, reducing the translation cost between what needs to be done and how to delegate it. 17. The episode touches on the open source contribution dynamic: TL Draw is 'source available' rather than MIT licensed, and Steve notes that the influx of AI-generated pull requests has forced changes to how the team participates in that contribution ecosystem. 18. The conversation style is informal and exploratory, with both host and guest working through half-formed observations in real time — neither is presenting a rehearsed thesis, which gives the episode an authentic quality but also makes it discursive. 19. Founders, CTOs, and senior engineers at small product teams who are actively integrating AI coding agents into daily work will find the most direct value here. 20. Listeners looking for a strategic overview of the AI coding tools market or beginner-level coverage of how to start using AI in development would find this conversation too inside-baseball and experience-focused.

As heard by us

A practical T.L. Draw conversation about selling SDKs as AI agents reshape developer tooling.

The Changelog uses T.L. Draw as a concrete case study in selling developer infrastructure while AI agents change how software gets built and discovered.

Read the full review in PlayNext →

Why you'd press play

Selling developer tools when language models are proliferating and one of them might be your best salesperson.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from The Changelog: Software Development, Open Source