Prompt your way to unicorn status

gm legends, happy Wednesday.

Xcode 26.3 is baking AI agents straight into the main tool for shipping iOS apps, Scribeist v2 turns what used to be a single blog tool into three focused spaces for novels, blogs, and everyday writing, and Universal-3 Pro lets you tell a speech model exactly how you want transcripts to look instead of fixing them with a maze of cleanup scripts.

Xcode 26.3 adds built-in support for agentic coding, so you can run coding agents like Anthropic’s Claude Agent and OpenAI’s Codex directly inside the IDE. They can break down tasks, change code, tweak project settings, run tests, and iterate until things compile, with support for the open Model Context Protocol if you want to plug in other tools.

🔥 Our Take: When AI coding shows up in Xcode, it stops being a side experiment and becomes part of the default path to the App Store. This is one of the main gates to iOS, macOS, watchOS, and visionOS, and now it comes with agents baked in. That means a lot of future apps will be touched by this by default, whether teams planned for it or not.

Geetanjali opened a thread asking founders and operators to zoom out from the usual “Silicon Valley or bust” debate and talk about where the strongest startup ecosystems are outside the US.

They’ve worked in India and France and seen how talent density, cost to build, access to capital, decision speed, and appetite for risk all shift by country. Now they’re asking: which ecosystem really punches above its weight, where talent, capital, and customers line up best, and which cities or countries are especially strong for certain stages (zero to one vs scaling) or verticals like AI, fintech, climate, SaaS, or deep tech. 

Scribeist v2 takes what started as a blog tool and turns it into a full writing platform. You get three workspaces: Novel for characters, timelines, visual canvases, and world-building; Blog for SEO-aware drafting, readability checks, research, and publishing; and General for distraction-free notes, lists, and everyday writing. Each workspace has its own structure, research tools, and AI tuned to that type of project.

🔥 Our Take: A lot of writers are quietly juggling Scrivener, Google Docs, planning boards, and a separate AI tab just to move one project forward. Splitting the work into novel, blog, and general spaces matches how people actually write instead of forcing everything into one catch-all editor. The real benefit is less friction, fewer open apps, and a lower barrier to just opening the right space and continuing.

Universal-3 Pro is AssemblyAI’s new speech model that lets you control transcription with a text prompt instead of building a long cleanup pipeline. You send audio plus instructions about domain terms, formatting, verbatim or cleaned-up output, tags, and speaker roles, and the model adapts in one pass. It is optimized for voice AI use cases, supports six languages, and is free to try through February for up to 5,000 hours.

🔥 Our Take: A lot of teams are quietly propping up transcription with extra LLM steps and fragile regex just to get something usable. Being able to push vocab, style, and tagging rules straight into the model cuts out a chunk of that glue work and makes the output easier to trust. If it stays accurate on messy real world audio instead of just clean samples, ripping out those post processing hacks becomes a very easy decision.