Turn merged features into a weekly changelog
Hey, prototype people, keeping customers informed about what you ship doesn't have to mean writing every update from scratch. Today’s workflow uses your GitHub history to draft a changelog, and we’ve got a new option for making decisions inside AI apps.
In today’s edition
- Google launches a multimodal decision API.
- Give your product three different AI reviewers.
Give your app a cheaper way to make choices
Google’s Gemini Decisions API scores predefined choices instead of writing a full response. You can use it to route requests, check policies, or send uncertain cases for review.
- One request can mix up to 1,000 yes-or-no questions, multiple-choice questions, and rubric scores. It accepts text, images, audio, video, and PDFs.
- The price is $0.10 per million input tokens, with no output-token charge. Google says its shared benchmark comparisons cost about one-fifth as much as Gemini 3.5 Flash-Lite through the Interactions API.
- It’s available in Google AI Studio and through the Gemini API, with a free tier for prototyping.
What it means Try it if your app needs to sort requests or check content against fixed criteria. Use the returned confidence scores to route uncertain cases for review.
Source: sundarpichai
Review your product from three angles
Allie K. Miller shares a repeatable way to get more critical feedback on a product or plan: ask separate AI reviewers to approach it from different perspectives, then pass their findings back for another review.
- In Claude, she assigns three subagents the roles of an enthusiastic user, a hostile critic, and a confused newcomer. You can also add descriptions of your ideal buyers.
- After the main agent collects the feedback, she sends it back to the reviewers for another pass. Ask for the complete list and which points were added or removed between rounds.
- She also suggests having a different AI system review the work, such as asking Codex to review something built in Claude.
What it means Try this before you ship a product change or finalize a plan. Start with the three reviewer roles and compare what each one catches.
Source: alliekmiller
Quick hits
- Google DeepMind released AlphaGenome 2, a DNA model that predicts variant effects in specific cell types, available through an API for noncommercial research and in private preview for commercial research.
- AWS released a free local emulator for testing supported cloud API calls without an AWS account; its AI agents wrote the code and help test and fix mismatches.
- Perplexity’s sandbox testing found that asking agents to find a hidden credential often drew them away from testing the virtual machine’s isolation.
- OpenRouter says its new routing algorithm cut the time to the first token by 40% and the price per million tokens by 17% for GLM 5.3 Flash users over the past month.
Claude Duet Review: is this the best AI agent yet?
See Claude Duet set up from scratch and compare its shared memory, channels, and tasks with other AI assistants. The review also covers its Anthropic-only model limitation and research-preview waitlist.
Watch on YouTube →AI workflow of the day
How to draft a weekly product changelog
Shared by Noah Kagan (@noahkagan) · View original post ↗
If you ship product updates but struggle to tell customers what changed, this prompt turns merged GitHub work into a dated changelog. Noah Kagan says TidyCal’s changelog page went from 60 to 3,437 Google impressions in three weeks.
- Connect GitHub to Claude Code, then open the project with your changelog.
- Paste the prompt below into Claude Code. Run it each Friday to add customer-facing changes to /changelog.
|
Every Friday, check everything we merged on GitHub. Skip internal fixes. For each new feature or change a customer would notice, write one line in plain words: what changed, why it's valuable to them, and a link to it if there is one. Put today's date on top. Add the entry to /changelog, newest on top.
|