What changed in AI, reviewed

Updates

Every AI change we have checked, newest first and in plain words: what changed, who it is for, and whether to try it, test it or leave it. The Weekly is different: one Monday email that picks three things worth trying.

46 updates reviewedNewest announced 21 October 2026Latest reviewed 23 October 2026

Start with the work impact, not the launch date.

Choose the job first

Are you catching up, or checking a specific tool?

Catch up on what matters

Stay on For most people. Start with the newest reviewed change, then use the verdicts to decide what is worth trying, testing or ignoring.

Find a provider or tool change

Open Filter, choose the maker, then narrow by verdict. The list stays an archive, not a ranking of providers.

Show

Verdict

Maker

Updates cover the wider market. Tool guides focus on ChatGPT, Claude, Gemini, Kimi and Grok.

NewestTest carefully

ClaudeAnnounced 21 October 2026Reviewed 22 October 2026

Claude Duet can prepare your morning brief. Check eligibility before relying on it.

Worth testing for people on US Pro and Max plans considering a low-risk brief or draft workflow. Prove it on one task before changing a live workflow.

Anthropic has opened a waitlisted US Pro and Max research preview for Claude Duet. It describes an app that watches connected work and prepares briefs and drafts; The Human Bit has not tested it.

Read the full update

Test carefullyChatGPTAnnounced 29 September 2026Reviewed 12 October 2026

OpenAI DevDay made agents, plugins and shared workspaces the new default. Start with one reviewed workflow, not the whole platform.

Worth testing for small teams using ChatGPT or Codex who now have more agent and collaboration surfaces than they can safely trial at once. Prove it on one task before changing a live workflow.

OpenAI's 29 September 2026 DevDay recap announced more than 20 changes across ChatGPT, Codex, the API and business plans: always-on Dots, GPT-6.1 Sol, Ultrafast, Private Intelligence, Codex Cloud, Agents API computer use, Decisions API, plugin extensions, MCP Events, ChatGPT Space, Pages, team tasks, ChatGPT in Slack and Microsoft Teams, Meetings, partner usage and the new Pro 500 tier. The useful move is not to try everything; it is to pick one workflow where AI can prepare the work and a person still checks the decision.

Test carefullyChatGPTAnnounced 22 September 2026Reviewed 22 October 2026

GPT-6 Sol and Luna lowered API prices. Check the model and the result before switching.

Worth testing for anyone running everyday ChatGPT Work or Codex tasks on GPT-5.6 Sol or Luna today. Prove it on one task before changing a live workflow.

OpenAI launched GPT-6 Sol and Luna on 22 September 2026 at roughly half their predecessors' promotional API rates. They are available in ChatGPT Work, Codex and the API, subject to plan, client and workspace availability. GPT-6.1 Sol is now a newer Sol option; this record describes the original GPT-6 models.

Test carefullyClaudeAnnounced 22 September 2026Reviewed 9 October 2026

Claude Opus 5.5 costs 40% less than Opus 5 and Anthropic says it matches Fable 5.1. Four things break on the way.

Worth testing for teams running an existing Claude Opus 5 or Opus 4.8 coding or agentic workflow. Prove it on one task before changing a live workflow.

Anthropic released Claude Opus 5.5 on 22 September 2026 as its current model for long-running agentic coding and knowledge work, at $4/$20 per million tokens against Opus 5's $5/$25. Anthropic says it performs at Claude Fable 5.1's level on most work while costing 40% less than Opus 5 to run. Opus 5 is now listed among Anthropic's legacy models, and four changes break code written for it, the largest being that thinking can no longer be disabled.

Test carefullyChatGPTAnnounced 21 September 2026Reviewed 22 October 2026

OpenAI is ending new self-serve fine-tuning jobs. Existing models have separate deadlines.

Worth testing for anyone who has built saved prompts, assistants or routines on a specific model. Prove it on one task before changing a live workflow.

New self-serve fine-tuning jobs end for active existing customers on 6 January 2027, not inference on every fine-tuned model. Evals, Agent Builder and reusable prompts have separate November shutdowns. Check the exact model and platform before planning a move.

Test carefullyGrokAnnounced 21 September 2026Reviewed 22 October 2026

Grok 4.7: test whether its self-checking improves your longer tasks.

Worth testing for teams already running a Grok 4.6 coding or agentic workflow. Prove it on one task before changing a live workflow.

xAI released Grok 4.7 on 21 September 2026 as the model it now recommends for code and chat, built on a new, larger base model with a longer reinforcement-learning run weighted toward problems that take hours to complete. Access, headline pricing and the context window match Grok 4.6, so the case for moving rests on whether the deeper reasoning and self-verification claims hold on your own long-running task.

Showing 6 of 34 reviewed updates.

The Human Bit rule

Mainstream news tells you what launched. We tell you what changed in the work.

Every update should leave you knowing whether to try, test, wait or ignore, and which judgement still belongs to the person responsible for the result.