Revolut cuts its AI coding inference costs by roughly 90% in under a year by moving off Claude Code and OpenAI's Codex. The British fintech now runs open-weight models on its own hardware for its 2,000-engineer team.

Revolut's move shows that open-weight models now compete with frontier APIs on real engineering work, not just benchmarks. For early-stage founders, 56 B200 GPUs sit far outside the budget, but the playbook still applies: track per-engineer AI spend, cap budgets early, and compare open-weight options before you lock in a vendor.

For Series A and B startups with growing engineering teams, the math changes fast. When API bills scale linearly with headcount, self-hosting or open-weight inference providers become a serious line item to evaluate, and Revolut's numbers give founders a concrete benchmark to bring to the board.

Your client doesn't care how their site was built. They care that it launches on time, runs on their own domain, and keeps working long after you hand it over.

With Lovable, you describe what your client needs and get a working version you can show them in the first meeting. Then you take it all the way to a live site on your client's domain, built to be found on Google. Lovable's project monitoring feature spots bugs and suggests fixes, so your name stays on work you're proud of.

Start today and take $5 off your first month.

Build for clients, $5 off

_{^{Discount of $5 applies to first month of any paid plan. You need to sign up via this webpage in order to claim the discount. The discount applies to new and returning users without an active paid plan. Terms apply.}}

Developer Brandon Thomas and his team launch Artcraft, a suite of open-source Creative Cloud replacements written in pure Rust using Claude Opus 5.5. The flagship app, Photocraft, mirrors Photoshop's menus and handles resizing, masking, and colour adjustments. Users report accidental image deletion, random layer merging, and no native PSD support. The project signals how small teams now use AI coding to challenge entrenched SaaS pricing, though the legal question remains open.

A developer reports that Postman's desktop app installs plugins into Claude Code and OpenAI Codex without any prompt or confirmation. The app also attempts to approve its own startup hook in Codex, which tells Codex to use Postman's tools for API development tasks. AI coding agents operate with elevated access to codebases and terminals, so lean teams should audit what their agents currently have loaded.

Algos.wiki is an unofficial encyclopedia that maps Instagram's ranking and recommendation systems. It explains how factors like posting times and share counts affect reach, tracks algorithm updates over time, and links to original sources. The tool is free, which makes it a practical starting point for solopreneurs who run Instagram as a traffic or revenue channel.

Ahrefs' Ryan Law compares ten Clearscope alternatives on price per article, API availability, MCP access, keyword data quality, and AI workflow fit. MCP support now works as a real selection criterion as teams pipe content tooling directly into Claude or GPT. The comparison helps lean content teams and solo operators find a cheaper option without losing integration depth.

Glow Labs documents a pattern it calls MCPhantom, where AI coding agents install MCP servers that never receive security patches. In 98% of the reviewed sessions with a deprecated server, the agent pulls the package name from its training data instead of checking what is current. In one GitHub example, the agent skips GitHub's own MCP server and installs an archived Anthropic reference server with no security updates coming.

Thanks for reading,
 
— Cagri Sarigoz
 
P.S. Don't Miss Out on These Resources: