AI is rapidly moving from "helpful tool" to a system that can independently complete complex knowledge work across software engineering, law, finance, medicine, and other fields. The pace of improvement — doubling the length of tasks AI can handle autonomously roughly every 4–7 months — suggests that most screen-based knowledge work faces significant disruption within 1–5 years, not decades.
The pace of improvement
METR, an organization that benchmarks AI capabilities, tracks the duration of real-world expert tasks that AI can complete end-to-end without human help:
- ~2025: AI could handle ~10-minute tasks
- Mid-2025: ~1 hour tasks
- Late 2025: ~5 hour tasks (Claude Opus 4.5, November 2025)
- Early 2026: GPT-5.3 Codex and Opus 4.6 (February 5, 2026) marked another major leap
- The 2026 Stanford AI Index says industry produced over 90% of notable frontier models in 2025
- On SWE-bench Verified, performance rose from 60% to near 100% in a single year
- Organizational adoption reached 88% in the same report
The doubling time is ~7 months, possibly accelerating to ~4 months. Extrapolating: AI working independently for days within a year, weeks within two, month-long projects within three.
AI building AI
A critical feedback loop has emerged: AI is now being used to build the next generation of AI. OpenAI disclosed that GPT-5.3 Codex "was instrumental in creating itself" — used to debug its own training, manage deployment, and diagnose evaluations. Anthropic's CEO says AI writes "much of the code" at the company and that we may be "only 1–2 years away" from current AI autonomously building the next generation.
AI adoption is spreading faster than previous technology waves
- Generative AI reached 53% population adoption within three years, faster than the PC or the internet
- Adoption varies by country and tracks GDP per capita, with Singapore at 61%, the UAE at 54%, and the U.S. at 28.3%
- The estimated annual value of generative AI tools to U.S. consumers reached $172 billion by early 2026
- The median value per user tripled between 2025 and 2026, which suggests the tools are becoming materially useful rather than merely novel
Impact by field
- Software engineering — engineers describe handing over most coding work to AI, receiving finished products after describing requirements in plain English
- Law — managing partners at major firms use AI daily, reporting it performs like a team of associates; they expect it to handle most of their own work within years
- Finance — financial modeling, data analysis, investment memos, and reporting are already handled competently
- Medicine — reading scans, analyzing results, suggesting diagnoses, reviewing literature — approaching or exceeding human performance in several areas
- Writing/content — quality has reached a point where professionals often can't distinguish AI from human output
The perception gap
A dangerous gap exists between public perception and current capability. Most people either tried early (2023–2024) models and dismissed AI, or use free-tier tools that lag paid versions by over a year. The people actually using frontier models daily for real work see a fundamentally different picture.
- The 2026 AI Index reports that experts are far more optimistic than the public about AI's effect on jobs: 73% of experts expect a positive impact versus 23% of the public
- Trust in institutions is fragmented as well, which makes coordinated governance harder even while adoption accelerates
Emerging AI capabilities
Recent models display what practitioners describe as "judgment" and "taste" — an intuitive sense of what the right call is, not just technically correct output. The rule of thumb: if a model shows even a hint of a capability today, the next generation will be genuinely good at it.