The AI Friday Brief
Good Thursday, NOLA. The Fable 5 story just got bigger: free week of access is live, and the internet has thoughts on whether this "relaunch" is real or marketing spin. We're also tracking a $30M bet to take on Microsoft Office with AI, Meta capping its internal AI spending, and some solid takes on whether AI is actually making us faster or just making us *feel* faster.
Models & Access
Fable 5 is free for a week — here's what to actually try
Yesterday's story about Claude Sonnet 5 launched alongside the return of Fable 5 and Mythos 5 (after export controls were lifted). The free access week is now live. The internet's response is split: some builders are genuinely excited, others are asking whether the relaunch is actually new or just a smart marketing reset after the export drama.
Anthropic
Kimi K2.7 Code now in GitHub Copilot
One more option in Copilot's model lineup: Kimi K2.7 Code is available for testing. If you're already in Copilot, you can switch to it in settings. Worth trying if you've heard good things about Kimi's coding performance.
GitHub
Industry Moves & Business
Indian tech founder bets $30M to build an AI alternative to Microsoft Office
Bhavin Turakhia (founder of Zeta, Instamojo, and others) is backing Neo, a new startup taking on Microsoft Office and Google Apps with AI as the core differentiator. His bet: AI-native productivity beats bolt-on AI. This is his fifth venture, and he's putting real money where his mouth is. Interesting timing when everyone's questioning whether specialized AI models can compete with frontier LLMs.
TechCrunch
Meta is capping internal AI token spending to control costs
As Meta's internal AI usage costs have approached billions this year, they've implemented spending caps and quotas. This is a real signal: even companies with enormous cloud infrastructure are feeling the pain of LLM inference costs at scale. Worth watching as more enterprises hit the same wall.
MLQ
Developer Tools & Platforms
Claude Desktop now available on Linux (beta)
If you've been waiting to run Claude Desktop natively on Linux, the beta is live. One of the big friction points for Linux-first developers is now resolved. Still beta, but worth testing if that's your workflow.
Anthropic
Godot game engine halts AI-authored code contributions
The Godot Foundation announced they'll no longer accept pull requests written entirely by AI. The reasoning: maintainers need to trust that contributors understand what they're submitting. This is a pragmatic call for an open-source project, not a philosophical stance against AI. Expect more OSS projects to make similar decisions as the volume of AI-generated PRs grows.
PC Gamer
On Speed, Cost, and Perception
"The gauge broke": Do AI tools actually make us faster?
A developer reports feeling 20% faster with AI but measuring 19% slower completion times. The disconnect is real: you *feel* more productive because you're not waiting on context switches, but the total time-to-done doesn't always match the feeling. Smart take on why developer velocity metrics are tricky in the AI era.
Hacker News
How Sonnet 5 compares across benchmarks — and why the pricing math is trickier than it looks
Sonnet 5 ranks #5 on the intelligence index but costs more per task than Opus because it generates nearly 2x the tokens. The real cost isn't just about model capability — it's about how efficiently the model solves your specific problem. Worth revisiting your cost assumptions if you were planning to upgrade.
AI Daily Brief
Also
- Sonnet 5 helped someone build 5 web apps in 10 minutes — Real-world Sonnet 5 usage
- OpenAI discovers new way to cut inference costs in half
- DeepSeek open-sources DSpark framework for 85% inference speedup — Open-source inference optimization
- Use Fable 5 as the superintelligent advisor, Sonnet 5 as the fast implementer — Workflow strategy for the new models
- After ~1B tokens with Sonnet 5: it's genuinely good if you use it very differently — Real usage insights on Sonnet 5's strengths
- Anthropic preps Claude agent for Microsoft Teams — Claude is moving into enterprise collaboration platforms
- AWS creates Forward Deployed Engineer organization — AWS building the forward-deployed engineer playbook
- CursorBench: Sonnet 5 is a meaningful step up from 4.6, middle of the pack overall — Coding benchmark analysis
- Why We're Building the Model Behind Base44 — Vibe-coding platform building their own model
- AI Engineer World's Fair hosting full FDE mini-conference this week — Forward-deployed engineers talking shop this week
Pass it on