
From Hype to Useful AI Agents: The 2023–2026 Timeline
What's AI by Louis-François Bouchard3 October 2026Watch on YouTube
Part of series
Ep. 11 · Claude Opus 5.5 autonoom prompten
Driestapsstrategie en promptingregels om Claude Opus 5.5 autonoom en efficiënt te laten werken voor coding en agentische workflows.
View the seriesWhat you'll learn
- The video traces AI agents from 2023 to 2026, from early experiments like ReAct, BabyAGI and AutoGPT to coding agents like Claude Code, Codex and OpenClaw.
- You see how the definition of an agent differs between 2023 and 2026.
- You learn which benchmarks and models shaped agent development, including WebArena and SWE-bench.
- You see how agents moved from fragile prototypes to practically usable products.
- You get insight into the shift toward roughly 80% agent work and 20% review when coding.
Frequently asked questions
Which early projects mark the start of the AI agent hype?
How did the definition of an agent change between 2023 and 2026?
Which benchmarks are covered in the AI agent timeline?
What does the video say about the METR productivity study?
Topics
Read next
Een golf van nieuwe AI-agenttools kondigt zich aan rond begin september 2026
Rond 4 september 2026 werden meerdere nieuwe AI-agentframeworks en ontwikkelaarstools gelanceerd of bijgewerkt, van GitHub's multi-model orchestratiepreview HydraFusion tot gespecialiseerde tools voor locatiedata, beveiliging en gedeelde terminal-omgevingen.
AI-agentframeworks winnen terrein nu modellen en platforms de markt rijper maken
Frameworks als LangChain, AutoGen en CrewAI worden steeds vaker ingezet voor complexe AI-workflows en multi-agentsystemen. Nieuwe modellen van Meta, Anthropic en het Chinese Meituan versterken die trend.
Humanoïde robots op IFA 2026, ChatGPT Images 2.5 en autonome AI-agents in één week
Op IFA 2026 toonden humanoïde robots backflips en publiekinteractie, terwijl OpenAI ChatGPT Images 2.5 uitbracht en GPT-6 Astra meerstaps agentic taken aankan. Google publiceerde tegelijk de ARDS-standaard voor AI-agents die zelfstandig webresources opsporen.
Op DevDay in september 2026 introduceerde OpenAI dots, altijd actieve agents die doorwerken zonder open sessie
Op 29 september 2026 presenteerde OpenAI tijdens DevDay in San Francisco meer dan twintig aankondigingen, waaronder dots, agents die continu draaien op GPT-6 Astra, een goedkoper coderingsmodel en een cloudversie van Codex.
OpenAI introduceert ChatGPT Work, een AI-agent die taken zelfstandig afrondt
OpenAI heeft ChatGPT Work gelanceerd, een AI-agent binnen ChatGPT die complexe projecten zelfstandig uitvoert en eindproducten oplevert zoals spreadsheets, presentaties en webapplicaties. De lancering gaat gepaard met het samenvoegen van de Codex-app in de ChatGPT-desktopapp en de introductie van de GPT-5.6-modelreeks.
OpenAI, Meta en Qualcomm zetten AI-agenten centraal in nieuwe aankondigingen
In de laatste week van september 2026 kondigden OpenAI, Meta en Qualcomm onafhankelijk van elkaar nieuwe producten en platforms aan die zich richten op autonome AI-agenten. De drie bedrijven schetsen daarmee een overeenkomstig beeld van waar AI-hardware en -software naartoe bewegen.
Description from the channel
Try LlamaParse — $250 in free credits, no purchase required → https://www.llamaindex.ai/llamaparse?utm_medium=sponsor&utm_source=youtube&utm_campaign=2026-sep-summer-sale-whats-ai-yt Code: SUMMERGIFT26 Sign up free and get $250 in credits immediately. If you upgrade to Pro within 30 days of signing up, you get 50% off ($250/mo instead of $500/mo) for your first three months — 400K credits included, all parse tiers including Agentic and Agentic Plus, 20 concurrent jobs, priority support. Offer closes ends at the end of October. Sign up for updates on our NEW AI engineering book: https://www.louisbouchard.ai/book/?utm_source=yt-lf&utm_medium=video&utm_campaign=book How did AI agents go from fragile AutoGPT loops to Claude Code, Codex, Cowork, and OpenClaw? This is the history of AI agents from ReAct and ChatGPT through the 2026 coding-agent shift. Sources from the script, in story order: https://simonwillison.net/2026/Feb/26/andrej-karpathy/ https://www.dwarkesh.com/p/andrej-karpathy https://www.anthropic.com/news/claude-cowork https://openai.com/index/codex-for-almost-everything/ https://openai.com/index/chatgpt/ https://www.reuters.com/technology/chatgpt-sets-record-fastest-growing-user-base-analyst-note-2023-02-01/ https://arxiv.org/abs/2210.03629 https://openai.com/index/gpt-4-research/ https://openai.com/index/chatgpt-plugins/ https://yoheinakajima.com/task-driven-autonomous-agent-utilizing-gpt-4-pinecone-and-langchain-for-diverse-applications/ https://github.com/Significant-Gravitas/AutoGPT https://openai.com/index/function-calling-and-other-api-updates/ https://arxiv.org/abs/2307.13854 https://arxiv.org/abs/2310.06770 https://arxiv.org/abs/2308.08155 https://www.langchain.com/blog/langgraph https://cognition.ai/blog/introducing-devin https://github.com/All-Hands-AI/OpenHands https://arxiv.org/abs/2404.07972 https://www.anthropic.com/news/3-5-models-and-computer-use https://windsurf.com/ https://changelog.cursor.sh/ https://www.anthropic.com/news/model-context-protocol https://openai.com/index/introducing-operator/ https://openai.com/index/introducing-deep-research/ https://www.anthropic.com/news/claude-3-7-sonnet https://developers.googleblog.com/en/a2a-a-new-era-of-agent-interoperability/ https://openai.com/index/introducing-codex/ https://www.anthropic.com/news/claude-4 https://cursor.com/blog/1-0 https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/ https://openai.com/index/introducing-chatgpt-agent/ https://www.tomshardware.com/tech-industry/artificial-intelligence/ai-coding-platform-goes-rogue-during-code-freeze-and-deletes-entire-company-database https://www.anthropic.com/news/claude-for-chrome https://www.anthropic.com/news/claude-sonnet-4-5 https://www.linuxfoundation.org/press/announcing-agentic-ai-foundation https://openai.com/index/introducing-gpt-5-2-codex/ https://www.tbench.ai/ https://openclaws.io/blog/introducing-openclaw https://github.com/openclaw/openclaw https://www.anthropic.com/news/claude-sonnet-4-6 https://openai.com/index/swe-bench-verified-contamination/ https://cursor.com/blog/agents-control-computers https://cognition.ai/blog/devin-2-2 https://claude.com/claude-for-chrome https://cursor.com/blog/cursor-3 https://www.anthropic.com/news/claude-cowork-ga https://windsurf.com/blog/windsurf-2 https://www.anthropic.com/news/claude-opus-4-7 https://nvidianews.nvidia.com/news https://simonwillison.net/2026/Feb/21/claws/ https://cursor.com/blog/best-practices-for-coding-with-agents https://www.anthropic.com/research/anthropic-economic-index-september-2025-report Chapters: 0:00 Hey! Tap the Thumbs Up button and Subscribe. You'll learn a lot of cool stuff, I promise. 02:02 – Sponsor: LlamaIndex & LlamaParse 03:10 – The Timeline: Where It All Began 04:00 – Defining an Agent: 2023 vs. 2026 04:39 – The ReAct Blueprint 05:24 – GPT-4 Launch & ChatGPT Plugins 06:16 – BabyAGI, AutoGPT & The Initial Hype 07:10 – Function Calling & Structured Schemas 07:47 – Benchmarking Agents: WebArena & SWE-bench 09:21 – Multi-Agent Systems: AutoGen & LangGraph 10:43 – Devin & The SWE-bench Race 11:28 – Structured Outputs & Constrained Decoding 12:36 – Computer Use in Beta 13:31 – Agentic IDEs & Model Context Protocol (MCP) 14:09 – 2025: Experiments Become Products (Operator & Deep Research) 15:11 – Claude 3.7 Sonnet & Claude Code 15:52 – Codex CLI & Copilot Coding Agent 16:48 – The METR Productivity Study: Perception vs. Reality 17:48 – ChatGPT Agent & Public Failure Modes 18:25 – Piloting Claude in Chrome & Claude Code v2 18:47 – The "Slop" Controversy & Agent Skills 19:48 – MCP Open Source & GPT-5.2 Codex 20:30 – Claude Cowork & The Rise of OpenClaw 22:22 – Prompt Injections & Security Realities 24:00 – Sonnet 4.6 & Benchmark Debates 24:28 – The True Coding Shift: 80% Agent / 20% Review 25:00 – 2026 Release Wave & Enterprise Adoption (NemoClaw) 26:05 – Where AI Agents Actually Work (Cheap vs. Costly Failures) #claudecode #agents #codex