Loading interactive tool…

This page in plain text

The AI timeline: 26 launches that changed how people use AI

From the day ChatGPT opened to the agents of today. Each entry gives the date its maker published the announcement, what changed for someone using AI, and a link to the source. Last checked on 20 November 2026.

2022

  • 30 November 2022. OpenAI: ChatGPT opens to everyone. A chat box anyone could type into. For the first time, using AI meant writing a sentence, and how you asked started to matter. Source. Next: How to write a brief.

2023

  • 14 March 2023. OpenAI: GPT-4 raises the ceiling. A model good enough for real work: long documents, harder reasoning, and pictures as input. AI moved from a curiosity to a tool people paid for. Source. Next: Twelve jobs to hand to AI.
  • 14 March 2023. Anthropic: Claude arrives. A second serious assistant, on the same day as GPT-4. From here on, choosing a tool was a real choice. Source. Next: The tools we use.
  • 20 July 2023. OpenAI: Custom instructions make context stick. You could tell ChatGPT who you are and how you like answers once, and it would keep that in mind. The start of AI that remembers your context. Source. Next: What context a brief needs.
  • 6 November 2023. OpenAI: GPTs: a prompt becomes a reusable assistant. Anyone could package instructions and knowledge into a custom version of ChatGPT and share it. A good prompt stopped being something you retype. Source. Next: Briefs already written.
  • 6 December 2023. Google: Gemini joins the race. Google’s model family, built from the start to work with text, images, audio and video together, and soon built into Gmail and Docs. Source. Next: The tools we use.

2024

  • 4 March 2024. Anthropic: Claude 3 comes in three sizes. Opus, Sonnet and Haiku: a large, a middle and a small fast model. Picking the size that fits the job became part of using AI well. Source. Next: Which Claude model to use.
  • 13 May 2024. OpenAI: GPT-4o talks, listens and sees. One model that handles voice, vision and text in real time. You could hold a spoken conversation with AI and point your camera at a problem. Source. Next: Practise an interview out loud.
  • 21 June 2024. Anthropic: Claude 3.5 Sonnet and Artifacts. Claude began building things beside the chat: a working page, a document, a small tool you could see and change as you talked. Source. Next: Build without a dev team.
  • 12 September 2024. OpenAI: o1 thinks before it answers. A model that spends time reasoning through a problem before replying. Slower and better on hard questions, and the start of choosing how much thinking a task deserves. Source. Next: Which Claude model to use.
  • 22 October 2024. Anthropic: Computer use: AI operates the screen. An experimental ability for Claude to look at a screen, move the pointer, click and type. AI started doing the task, not just describing it. Source. Next: What a coding agent is.
  • 25 November 2024. Anthropic: The Model Context Protocol. An open standard for connecting assistants to the places your data lives: files, business tools, databases. The plumbing that lets an assistant work with your own material. Source. Next: Put AI inside your own site.
  • 11 December 2024. Google: Gemini 2.0 and the agentic era. Google framed its new models around agents: AI that plans several steps ahead and acts on your behalf, under your supervision. Source. Next: Twelve jobs to hand to AI.

2025

  • 2 February 2025. OpenAI: Deep research. Ask a question, walk away, and come back to a long report with sources. Research became something you could delegate, then check. Source. Next: Research a question and see the sources.
  • 24 February 2025. Anthropic: Claude 3.7 Sonnet and Claude Code. Claude Code arrived as a preview: an agent that works in your project from the command line, reading files, running commands and making changes. Source. Next: Claude Code and Codex.
  • 25 March 2025. OpenAI: Images inside the conversation. Image generation moved into ChatGPT itself, with readable text and the ability to change a picture by asking. Making an image became a conversation. Source. Next: Make an image for a post or deck.
  • 16 May 2025. OpenAI: Codex takes on coding tasks. OpenAI’s coding agent could work on several software tasks at once in the background. Two agents from two labs made “have one review the other” practical. Source. Next: Claude Code and Codex.
  • 20 May 2025. Google: Veo 3 adds sound to AI video. Generated video arrived with its own dialogue, effects and ambient sound. A short clip no longer needed a camera or a microphone. Source. Next: Record a voiceover without a studio.
  • 22 May 2025. Anthropic: Claude 4. Models built to keep working on a long task for hours, and Claude Code became generally available. Delegating a whole job, not a single step, came into reach. Source. Next: Let the builder write your brief.
  • 17 July 2025. OpenAI: ChatGPT agent. ChatGPT gained an agent that thinks and acts, using tools to complete tasks such as research and slideshows with your guidance. Source. Next: Twelve jobs to hand to AI.
  • 7 August 2025. OpenAI: GPT-5. One system that decides for itself when to answer quickly and when to think longer. Less choosing between models, more describing the job well. Source. Next: How to write a brief.
  • 16 October 2025. Anthropic: Agent Skills. Folders of instructions, scripts and resources that Claude loads only when a task calls for them. Know-how became something you write once and reuse. Source. Next: Free courses on building with AI.
  • 18 November 2025. Google: Gemini 3. Google’s next generation, across the Gemini app, Search and its developer tools. All three major labs now shipped models aimed at long, multi-step work. Source. Next: The tools we use.

2026

  • 30 June 2026. Anthropic: Claude Sonnet 5. The first of the Claude 5 models, described by Anthropic as its most agentic Sonnet yet, for coding and everyday professional work. Source. Next: Which Claude model to use.
  • 17 November 2026. Anthropic: Claude Opus 5.5 and Sonnet 5.5. Faster and cheaper to run than the models they replace. Sonnet 5.5 answers members’ questions on this site, and Haiku 5.5 answers guests’. Source. Next: Our test of three Claude models.
  • 17 November 2026. OpenAI: GPT-6.2 Astra and Sol. OpenAI’s newest models, announced the same day, aimed at professional work, coding and operating a computer at lower cost. Source. Next: The tools we use.

The pattern

Three shifts run through all of it. AI went from answering a question to doing the work while you do something else. Custom instructions, GPTs, connectors and skills all stopped you explaining yourself again every time. And the longer an AI works alone, the more depends on what you told it at the start, which is why we teach how to write a brief.

Ask Another AI Guy

I'm just Another AI Guy.

You're chatting with our fast model. Create a free account for the more thorough one, and more messages.