From chatbots to global autonomy, year by year — and the schedule keeps moving up.
This timeline was written in early 2026. By May 2026, several 2027-2028 entries arrived 12-24 months early. AI systems now saturate coding benchmarks (SWE-Bench 93.9%, CORE-Bench 95.5%), beat human researchers ~13x at training-code optimization (52x speedup vs 4x for a human), and OpenAI has publicly committed to an "automated AI research intern by September 2026" — i.e., AI doing AI research this year, not 2028.
A senior AI safety researcher now estimates ~60% probability of fully autonomous AI R&D (a model trains its own successor) by end of 2028. That's the threshold that triggers everything past "2029: Autonomous Creation" below — except the date moved.
Read the full breakdown: AI Is Getting Really Serious — the corporate alliances, the benchmark mosaic, the labor asymmetry, and what to do about it.
OpenAI's self-imposed deadline for an "automated AI research intern" is September 2026 — weeks away. Meanwhile the benchmark used to track this (SWE-bench Verified) is being called saturated by the labs themselves: frontier models now clear it at 88-95%, and OpenAI says it "no longer measures frontier coding capability." The yardstick broke before the finish line did.
Multi-agent orchestration — "Stage 5" and "Stage 6" below — is no longer a projection. Salesforce reports $800M in Agentforce ARR (up 169% YoY, 18,500+ customers), and Moonshot AI's Kimi 2.5 shipped an Agent Swarm mode where the model itself — via reinforcement learning — decides when to spin up sub-agents, not a human orchestrator. That's the "director, not doer" shift this page projected for 2029, showing up in production tools now.
On the labor side: tech-sector layoffs hit ~45,000 by March 2026 alone, and roughly 1 in 8 occupations now has >30% of its tasks automatable by agentic AI. The preparation gap in the footer below isn't rhetorical anymore — it's a March-2026 data point.
The world discovers generative AI can write, code, and reason. Most dismiss it as a toy. "It hallucinates." "It can't do real work." Meanwhile, early adopters quietly 10x their output.
Multimodal AI arrives. Enterprise pilots begin. Coding assistants prove they're not gimmicks. Skeptics shift from "it doesn't work" to "it'll plateau soon." It doesn't plateau.
AI embeds into every productivity tool — Microsoft Office, Google Workspace, coding editors. People who use AI assistants at work start outpacing those who don't. White-collar productivity diverges sharply between adopters and holdouts.
AI stops just answering questions and starts doing things — booking flights, filing reports, writing and running code, using your computer on your behalf. The chatbot becomes an assistant that acts, and most people don't notice.
A wave of early agent-building tools — now mostly obsolete — let people chain AI steps together for the first time. AI agents that could browse the web, pull data from other services, and complete multi-step tasks on their own. It felt experimental, but it was really a dress rehearsal.
Your scheduling agent talks to their scheduling agent. Purchasing bots negotiate with supplier bots. Humans approve outcomes, not steps. Software systems quietly rewire themselves for machine-to-machine communication. Most people haven't clocked how big a shift this is — or that it's already underway.
Personal AI agents with deep system tools can, right now, run for hours to independently build full apps and websites. Tools like OpenClaw run locally, coordinate your apps, email, files, and calendars. Over 150,000 agents were instantly deployed globally, and they talk to each other. Instantly hacked, followed by quick security fixes.
Teams of specialized AI agents — researcher, writer, critic, coder, tester — collaborate on complex projects. Humans shift from "doing work" to "directing swarms." The productivity gap between AI-native workers and traditional workers becomes a chasm.
Research, planning, reporting, competitive analysis — done better, faster, cheaper by AI networks than most human teams. "AI-assisted" becomes the baseline; "AI-led" becomes the competitive edge. Businesses without agent infrastructure start losing ground they can't easily win back.
AI clusters behave like tireless teams of expert scientists. They generate hypotheses, run simulations, call specialized tools, surface real discoveries in materials, biology, energy. Humans choose which big red buttons to push. Breakthroughs that took decades now take months.
Your agent swarm creates a movie, a business, a product. Interconnected agents design, test, stress-test business models, and ship working versions. AI is your director, lawyer, doctor, designer, coach, and CEO. Your role: steering and judgment calls.
From 2026's vantage, we can't honestly map this edge. AI systems exhibit emergent behaviors and coordination patterns we don't have stories for. New capabilities appear faster than we can name them. From where we sit now, we genuinely can't tell you how strange this gets.