Updates for 5/31

Salesforce rolled out Claude Code to its entire dev org with no token limits, cutting a 231-day internal migration to 13 days, with 79% more pull requests per developer and 5% fewer incidents reported for April 2026 (numbers not independently verified, The Decoder)

Until last month, AI coding tool announcements rarely came with concrete timelines or PR metrics — most stories topped out at 'teams are actively using it.' Full engineering-org rollouts at a Salesforce-scale company were practically uncharted territory. Today those numbers arrived: 231 days down to 13, 79% more PRs (no independent verification yet).

This is the benchmark engineering managers have been waiting for to justify a procurement ask. Organizations at similar scale now have a comparison point — or a target to beat. Small teams of a few dozen are unlikely to feel this directly; they're already iterating based on feel. The absence of independent verification is worth noting, but the magnitude of the claimed result is hard to ignore.

Internally adopted Anthropic's Claude Code across its dev org with no token limits, cutting a 231-day migration to 13 days and reporting 79% more PRs per developer and 5% fewer incidents for April 2026 (numbers not independently verified, The Decoder) — positioned as a flagship case for the agentic shift

Salesforce sells Agentforce to its customers, but what its internal engineering org actually runs on had never been spelled out. The assumption was 'they use their own stack.' Today's disclosure breaks that: Salesforce's developers are on Anthropic's Claude Code — an external model — not Agentforce. For a vendor whose pitch centers on AI-powered agents, that's an unusually candid admission.

For enterprises evaluating Salesforce's AI stack, this clarifies the layering: Agentforce handles business processes; Claude Code handles engineering workflows — complements, not competitors. Procurement teams at Salesforce shops will now likely factor in Claude Code licensing as a natural pairing. For Anthropic, it's a high-profile reference that will show up in sales decks quickly. If you're not in a Salesforce environment, this is background context.

Codex now runs on Windows 11 with 'Computer Use' — the AI can independently drive applications, test apps, and hunt for bugs, with users able to start and monitor tasks remotely from the ChatGPT mobile app while away from the PC (The Decoder)

For the past six months, Codex was primarily a terminal-and-browser coding assistant — capable, but you had to be at the machine to see what it was doing. Overnight autonomous runs weren't practical without keeping a session alive. Today's Windows 11 + Computer Use update changes the model: the AI operates the PC independently, and you check in from your phone.

'Let the AI hunt for bugs overnight' or 'kick off a test run while commuting' moves from hypothetical to realistic. You no longer need to be at the machine to start or monitor a task. The trade-off: autonomous PC operation raises the stakes for permission scope — giving the AI broad file access unsupervised needs careful sandboxing. If you stay in CLI or editor workflows, this is still more curiosity than daily tool.

ChatGPT sidebars added inside Microsoft Excel and Google Sheets, powered by the new default GPT-5.5 Instant — which OpenAI claims hallucinates 52.5% less than GPT-5.3 Instant on high-stakes prompts (medicine, law, finance)

Until last month, using ChatGPT with a spreadsheet meant copy-paste round trips — pull the data out, ask the question, paste the answer back. Excel had Copilot, but that required a higher-tier Microsoft 365 subscription, and Google Sheets AI was limited to Gemini. Users who wanted GPT-5.5 inside a spreadsheet had no direct path. Now there's a sidebar for both.

For anyone who lives in spreadsheets, the workflow gets shorter — formula explanations, data interpretation, and draft summaries without leaving the sheet. The 52.5% hallucination reduction is OpenAI's own claim, not independently verified, so high-stakes financial or legal data still needs human review. If you're not regularly in Excel or Sheets, today's update is background noise.