Key Takeaways
- Import AI 455 (Jack Clark, May 4, 2026) estimates 60%+ probability of fully autonomous AI R&D — where systems improve themselves — by end of 2028.
- Autonomous AI R&D would accelerate the pace of AI tool development beyond today's trajectory, potentially rendering current AI software obsolete faster than expected.
- For practitioners evaluating multi-year tool commitments in 2026, this forecast has one immediate implication: AI software selected today may be two generations behind by contract end.
- The forecast aligns with AI 2027 predictions. If correct, it compresses the innovation timeline and raises the case for software flexibility and vendor adaptability in contracts.
- Contingency planning should focus on vendor track record for rapid updates and data portability — not feature parity locked in for 3 years.
What is the autonomous AI R&D forecast and who is making it?
Jack Clark, Import AI newsletter author and co-founder of OpenAI, published a forecast in Import AI 455 (May 4, 2026) estimating a 60%+ probability that fully autonomous AI R&D — where AI systems autonomously build and improve their own successors without continuous human direction — could arrive by end of 2028.
Clark is not a fringe forecaster. Import AI is one of the most widely-read AI research newsletters. His credibility rests on consistent, careful technical analysis rather than hype. The forecast includes supporting evidence from Anthropic's automated alignment researchers, published in Import AI 454 (April 2026), which demonstrated AI systems outperforming human baselines on weak-to-strong supervision tasks. These are early signals, not promises. But they're coming from credible research institutions.
The distinction matters: Clark is not predicting general AI or AGI. He is predicting a specific technical capability — AI systems that autonomously conduct research and improve themselves in the lab. That capability would collapse the feedback loop between AI development and deployment, potentially accelerating progress in ways the industry has not yet experienced.
Why does a 2028 timeline matter for decisions being made in 2026?
Practitioners and firms are signing multi-year contracts with accounting AI tools, payroll platforms, and practice management systems in 2026. Many of these contracts lock in pricing and feature parity for three years. If autonomous AI R&D arrives in 2028, the AI landscape at the end of a three-year contract (2029) could look completely different from today.
The pace of AI tool improvement is already faster than most practitioners expect. A tool that ranks 85–95% accurate in 2026 may be 98%+ accurate by 2027 if development accelerates. Multiply that acceleration further if AI systems are autonomously researching and improving themselves. The gap between your locked-in tool and the available alternatives could widen significantly before the contract expires.
This is not an argument to avoid AI tool adoption. It is an argument to scrutinize contract terms and vendor responsiveness. The software you evaluate today may be adequate in 2029. But it may also be painfully behind.
How does this forecast compare to existing AI 2027 predictions?
The AI 2027 framework predicted that AI agents would infiltrate the CFO office and accounting workflows by 2026, and that regulatory AI use would accelerate in 2025–2026. Both predictions have already started to materialize: PwC + OpenAI CFO agent announcement (May 4), IRS AI audit selection policy (formalized February 2026), Google Workspace AI (May 2026). The trajectory has been roughly accurate.
The autonomous AI R&D forecast from Import AI 455 extends and accelerates the timeline. If AI 2027 said "AI tools infiltrate enterprise and regulatory functions by 2026," Import AI is saying "and those tools will improve themselves by 2028." The timeline compresses from a linear progression to an exponential one.
Neither forecast is certain. Forecasting AI capability timelines is notoriously difficult — progress can be surprising in both directions. But when multiple credible sources (AI 2027 framework, Import AI 455, Anthropic research) point toward accelerating timelines, it becomes prudent planning context rather than speculation.
What is one concrete planning implication for practitioners?
When evaluating vendor contracts in 2026, ask: "What is your roadmap for AI-driven improvement to this software, and how do you handle breaking changes when capabilities improve dramatically?" The answer should address:
- Vendor track record for rapid updates: Has this vendor shipped major capability updates within 6-12 months before? Or do they move slowly?
- Data portability: If the tool becomes outdated or another vendor's solution is significantly better in 2028 or 2029, can you extract your firm's data and workflow configurations to migrate easily?
- Pricing flexibility: Does the vendor adjust pricing based on automation value as capabilities improve, or are you locked into 3-year fixed pricing while their value proposition multiplies?
These questions apply whether or not autonomous AI R&D arrives by 2028. But if it does, they become critical differentiators between tools that remain useful and tools that age rapidly.
Why this forecast matters for strategic planning
If autonomous AI R&D arrives by 2028, the accounting software and practice management platforms that firms adopt in 2026 may become significantly less competitive within 24–30 months. The vendors who continue to improve rapidly will create a wide gap versus those who stall.
For practitioners, the implication is not to avoid AI tools — the opposite. It's to choose tools and vendors with defensibility built in: vendors with strong technical talent, rapid iteration cycles, and demonstrated ability to ship meaningful updates. Lock in the vendor's responsiveness, not the feature set.
