A loaded week on the GitHub and Anthropic side, plus a deep run of Everpure conference coverage on the data infrastructure beat. Here's what mattered.
GitHub kills the PAT requirement for agentic workflows
A quiet but meaningful change: GitHub Agentic Workflows can now run on the built-in GITHUB_TOKEN instead of requiring a personal access token. That matters more than it sounds. PATs are long-lived, often over-scoped, and easy to forget about — exactly the kind of credential you don't want sitting inside an autonomous workflow that's reading repo content and triggering CI/CD. The new token is short-lived and scoped to the specific run. The change also shifts AI usage billing from individual developer accounts to the organization when workflows run on org-owned repos, which makes cost tracking and governance considerably easier for platform teams.
Read more
Anthropic pauses the Claude Agent SDK billing change — for now
Anthropic pulled back a planned billing overhaul on June 15, the same day it was supposed to take effect. The original plan would have split Claude usage into two buckets: standard subscription limits for chat and the official CLI, and API-rate billing for everything else built on the Agent SDK. Right now, SDK usage still runs under the generous weekly caps tied to your subscription tier — which is a real advantage for teams building agents on Claude. Anthropic hasn't said when a revised plan will land, so treat this as delayed, not cancelled.
Read more
Claude Code gets live, shareable Artifacts
Anthropic extended its Artifacts feature into Claude Code this week, letting engineering teams view AI coding sessions as real-time, interactive pages rather than static logs. For teams already using Claude Code in pair or async workflows, this closes a real gap — it's now easier to share what an agent did and why, without screen recordings or copy-pasted terminal output.
Read more
Other GitHub news worth a skim
GitHub also shipped two infrastructure updates this week: self-hosted runner version management is getting more rigorous, and GitHub Enterprise Server 3.21 is now GA. And on a more practical note: GitHub is retiring its free AI model playground — worth checking if your team relies on it for quick model testing.
Moonshot AI's Kimi K2.7-Code goes after token efficiency
Most agentic coding model releases lead with raw capability. Kimi K2.7-Code leads with something more practical: getting more done per token, which matters a lot once you're running agents at scale and the bill starts to add up.
Read more
Why giving AI agents your access badge is "dangerous as hell"
Angad Narang, VP of Product at Everpure, doesn't soften this one: when an agent acts on your behalf, the easy answer — give it the same access you have — is also the wrong one. This piece digs into why broad access delegation to agents is a governance problem waiting to happen, and what a more deliberate access model looks like in practice.
Read more
Everpure Accelerate coverage: the data infrastructure problem behind enterprise AI
Several pieces this week came out of Everpure Accelerate in Las Vegas, all circling the same theme: enterprise AI keeps stalling not because of model quality, but because of the data underneath it.
Also from Coder Legion this week
The through-line this week: identity and access for agents is becoming its own discipline — whether it's GitHub tightening token scope or Everpure's Narang warning against giving agents a badge with no limits.
See you next Friday.
Developer Weekly Briefing is published every Friday on Coder Legion. Written by Tom Smith.