Meta's Muse, sqlite-utils 4.0, and OpenAI's banking push

Meta launches Muse image generator, sqlite-utils hits 4.0 with built-in migrations, and MUFG goes all-in on ChatGPT Enterprise.

The big picture

Quiet stretch — a mix of one genuinely useful open-source release, a corporate AI partnership announcement, and Meta entering yet another crowded image-gen market. Nothing earth-shattering, but the sqlite-utils 4.0 release is the kind of practical tooling upgrade that actually improves your week.

sqlite-utils 4.0 absorbs sqlite-migrate (and that’s a good thing)

Simon Willison shipped sqlite-utils 4.0, a major version bump that folds schema migration support directly into the library. The companion library sqlite-migrate 0.2 is now essentially retired — it ships as a thin compatibility shim pointing at the new sqlite-utils internals rather than maintaining its own logic. Simon Willison’s Weblog

This is a smart consolidation. If you’ve been using sqlite-utils for scripting and data wrangling, you previously had to reach for a separate tool any time you needed to evolve a schema without blowing away data. Having migrations baked in means one fewer dependency and one fewer mental context switch. Worth noting: the 4.0 release candidate credits a detailed review by Claude Fable 5 for shaping the final API — which is an interesting data point on AI-assisted library design becoming a normal part of a solo developer’s workflow. sqlite-migrate 0.2

If you use sqlite-utils in any capacity, upgrade. If you were holding off on adopting migrations because sqlite-migrate felt like overkill for a side project, 4.0 removes that excuse.

Simon Willison’s other experiment: GPT-5.5 building a Web Component in one prompt

On the same day as the sqlite-utils release, Willison posted a small experiment: he gave GPT-5.5 a single prompt describing a Web Component that embeds GitHub code snippets inline, and published the result as a working tool. The component takes a GitHub file URL with a line range, fetches the raw content, and renders the specified lines with numbers. Simon Willison’s Weblog

There’s no syntax highlighting, and it’s explicitly labeled experimental, but the real story here is the workflow: a useful, shareable tool from a single natural-language prompt. Willison does this kind of thing regularly and it’s a good calibration for where GPT-5.5 actually sits on practical frontend tasks. The result is genuinely usable — not polished, but functional enough to embed in a blog post. If you’re skeptical that current models can handle real component-level tasks, this is worth a look.

Meta Muse enters a very crowded image-gen market

Meta announced Muse, a new AI image generator targeting advertising, home decor, and creator use cases. The announcement is thin on technical specifics — no architecture details, no benchmark comparisons, no API docs in the announcement. TechCrunch

Honestly, the launch reads more like a product marketing beat than a developer-relevant release. Meta is already behind Flux, Midjourney, and Stable Diffusion in mindshare among developers who actually build image-gen features. Unless Muse ships with a usable API and competitive output quality, it’s hard to care yet. Watch for whether this gets an actual developer-accessible endpoint before forming an opinion.

Enterprise AI deals: MUFG and OpenAI go deep on banking workflows

Japan’s MUFG (one of the world’s largest banks) announced a push to become what it’s calling an “AI-native” organization using ChatGPT Enterprise, targeting internal workflow improvements and new AI-powered financial services. OpenAI

This is squarely in the category of enterprise case study content — OpenAI publishing a partner success story. That said, the scale matters: MUFG has over 40,000 employees and operates across dozens of countries. Deployments at this scale are worth watching because the failure modes and compliance requirements are genuinely harder than a 200-person startup adopting Copilot. What’s missing from the announcement is any specifics on what workflows changed, what the rollout looked like technically, or how they’re handling financial data in prompts. The abstract is interesting; the substance isn’t published yet.

Quick hits

sqlite-utils 4.0rc4 was the final release candidate before stable, shaped largely by API feedback from a Claude Fable 5 code review — an unusually transparent look at AI-in-the-loop library development. Simon Willison’s Weblog