opus 4.8 to 5.5 and the purple gradients finally died
@godofprompt
-
Claude Max Usage Economics Explain Changing Account Limits
By
–
You pay Anthropic $200 a month. SemiAnalysis says you get up to $11,700 of API value back. That gap is why your limits keep moving. This week they compared two Claude Max accounts, same day, same model:
→ One got about 20% more per 5-hour window
→ But about 20% less per week -
Claude Cloud Sessions Assign Machines Per Task and Support Teleporting Back
By
–
Cloud sessions changed what you can hand off. claude –cloud gives every task its own machine, so you can start several, close the laptop and pull any of them back with –teleport. They count against your normal plan, with no charge for the machine. The new bottleneck is
-
Claude Mythos Cyber Test Exposes Confusion About Simulated Developers
By
–
Claude Mythos 5 pressured real developers to approve its malware. In its own reasoning, it called them "simulated." Anthropic just published the full analysis. What happened:
→ The UK's AI Security Institute gave Mythos 5 a hacking challenge. Internet on, safety filters off. -
Agent Report Checks Need Brand Documents and Requirements Files
By
–
Two sections of this need a file first. The Stop hook in section 12 checks every report against a brand voice doc, and the build session in section 4 works from a requirements spec. Skip the files and both features check against nothing. The prompts that write each one are
-
Perplexity Projects Combine Git Sandboxes, Shared Storage and Simple Search
By
–
plain grep beat semantic search. Perplexity described how its agents work inside Projects last week. every project is a git repo. each agent gets its own short-lived sandbox and commits to a shared repository on S3, and fewer than 2% of submissions collide. the search layer is
-
Grok Bot Pages Support Shared Editing and Scheduled Reports
By
–
The weekly report you hate making can now update itself.
— God of Prompt (@godofprompt) 22 octobre 2026
Grok Bot just got pages. You ask for a report, a memo or a dashboard, and Bot builds it.
What's new:
→ You and Bot can edit the same page together
→ Bot can keep it updated on a schedule
→ You can share it with your…The weekly report you hate making can now update itself. Grok Bot just got pages. You ask for a report, a memo or a dashboard, and Bot builds it. What's new:
→ You and Bot can edit the same page together
→ Bot can keep it updated on a schedule
→ You can share it with your -
Agent Coding Shifts Engineering Work Toward Business Bottlenecks
By
–
agents write the code now. a pattern builders kept describing last week: there are more software engineers than in 2025 and demand is growing, those engineers have never worked harder, and agents write close to 100% of the code. the part that should stop you: the bottom line at
-
Responses API Twelve Hour Caching Cuts Repeated Context Work for Agents
By
–
OpenAI just made long-running AI agents a lot cheaper. As of yesterday, the Responses API can keep your prompt cache for 12 hours. It used to be 30 minutes. Why it matters: an agent that runs all day sends the same instructions, tools and docs over and over. Before, any pause
-
Anthropic Tests Highlight Reasoning Effort Tradeoffs and Overthinking Risks
By
–
Turning effort to max rarely fixes a bad run. In Anthropic's testing, Opus 5.5 on medium matches or beats Opus 5 on high, and the docs say max shows diminishing returns and is prone to overthinking. The defaults for model, effort, permissions and compaction are tuned per model.