AI 2027 Turns One: The Boring Half Came True First

What happened to the AI 2027 forecast one year in?

The AI 2027 forecast split in two: frontier predictions like superhuman coding agents are running late, while business predictions like embedded accounting agents arrived early.

Fifteen months ago, a team of AI forecasters led by Daniel Kokotajlo published a scenario called AI 2027. It predicted a fast, dramatic path toward advanced AI: opaque reasoning models by January 2027, a stolen-weights cyberattack by February, a superhuman coding agent by March. We translated that scenario for business owners in April and told you what it meant for your software bills, your bookkeeping and your hiring plans.

There's now enough runway to check the forecast against reality. The answer isn't a simple hit or miss. It's a split. The parts aimed at frontier AI labs are landing late. The parts aimed at your business are landing early.

What do the AI 2027 authors say about their own timeline?

Kokotajlo says progress is going "somewhat slower" than predicted, and pushed his median timeline to around 2030, while co-author Eli Lifland has moved further out still, to January 2035.

Kokotajlo has said it himself. "Things seem to be going somewhat slower than the AI 2027 scenario," he wrote in a public post late last year. He moved his own median estimate for advanced AI from 2027 to around the end of 2030. Co-author Eli Lifland moved further out still. In a January 2026 update on their shared blog, Lifland put his median at January 2035, up from an April 2025 estimate of 2031.

A separate tracking assessment from FutureSearch dug into why. The scenario assumed frontier labs would pour their new AI agents into speeding up their own research, creating a feedback loop toward more powerful models. That hasn't happened at the pace the scenario needed. OpenAI leaned into consumer products instead, including its Sora app and shopping features. Meta focused on companion products and advertising. Only Anthropic showed a heavy focus on using coding agents to accelerate its own research.

Which AI 2027 predictions already came true for your business?

Embedded agents in accounting software arrived on schedule: QuickBooks Accounting Agent is live, Sage Intacct shipped agentic AI, and Intuit connected to Claude early.

Flip to the business side of the same forecast and the story reverses. The scenario predicted that by late 2026, agents cheap enough to embed directly into everyday software would show up inside accounting tools. That already happened.

Accounting agents are live, not promised

QuickBooks Accounting Agent now categorizes transactions using up to 24 months of history and a "policy memory" that remembers your rules. Upload a bank statement as a PDF or a photo and the agent extracts the transactions, compares them against your books and flags discrepancies. Sage answered with its own move. PwC and Sage now use agents to configure Intacct rollouts, and Sage's most recent release automated multi-entity consolidation, cutting into work that used to be a manual close-week grind.

We covered the most striking early move ourselves. In April, Intuit connected QuickBooks, TurboTax and Credit Karma directly inside Claude using Anthropic's Model Context Protocol. That was the forecast's late-2026 prediction landing months ahead of schedule.

The entry-level hiring squeeze is showing up in the data

The forecast also predicted job market disruption concentrated at the entry level, and that part is tracking too. A Stanford Digital Economy Lab study using ADP payroll data, led by Erik Brynjolfsson, found that workers aged 22 to 25 in the most AI-exposed occupations are seeing a real employment decline relative to their peers, with outlets reporting figures between 13 and 16 percent. Fortune's June follow-up put the annual decline for that age group at about 3.8 percent, up from 2.8 percent a year earlier, while workers 35 to 40 in comparable roles grew about 2 percent. The named occupations include software developers, customer service representatives and programmers: transaction-level work that looks a lot like entry-level bookkeeping.

Government projections point the same direction. The Bureau of Labor Statistics expects accountant and auditor employment to grow 5 percent from 2024 to 2034, with about 124,200 openings a year, mostly from people leaving the field. Bookkeeping and accounting clerk employment is projected to fall 6 percent over the same window, a decline the agency attributes to software automation. The shortage at the top of the profession and the shrinkage at the entry point are two sides of the same shift.

Nexairi Dispatch

Get the next AI move before it turns into a workflow problem.

Join the free newsletter for concise AI news, practical checklists, and the decisions practitioners need to make next.

Free. No spam. You will also get the Nexairi Dispatch.

The two-track scorecard

Business-layer predictions mostly arrived on time or early, while frontier-lab predictions, the ones with the dramatic headlines, are running behind by the authors' own admission.

Track	AI 2027 prediction	Status, July 2026
Business	Cheap agents embed inside accounting software by late 2026	Arrived early: QuickBooks and Sage Intacct agents are live
Business	Entry-level hiring softens as agents absorb routine work	Confirmed by Stanford/ADP data and BLS clerk-role projections
Frontier	Agent-2 with opaque "neuralese" reasoning by January 2027	Delayed: authors have pushed their own timelines back
Frontier	Superhuman coding agent (Agent-3) by March 2027	Delayed: R&D-acceleration precondition not met at labs

Why did the frontier predictions slip while the business predictions landed early?

Frontier predictions needed labs to redirect AI agents into research, and labs chose consumer products instead. Business predictions only needed agents good enough to categorize transactions.

The two tracks were never running the same race. The frontier half needed labs to point new AI capability at AI research itself, again and again, until progress compounded. That's a bet on lab incentives and technical difficulty stacking up in a narrow window. It hasn't happened yet, and the scenario's own authors say so.

The business half needed something easier: agents reliable enough to categorize a transaction or flag a mismatched invoice. That bar is low enough that Intuit, Sage and a handful of fintech vendors cleared it inside a normal product cycle. Nobody needed a research breakthrough. They needed a product team with a deadline.

Why this split matters more than either headline alone

It's tempting to read "frontier AI is running late" as a reason to relax. That misses the point. The predictions that touch your firm's daily work, software costs, categorization agents and hiring pressure on juniors, are the ones that came true. Some came true early. Superhuman coders and opaque reasoning models make for better headlines, but they were never the part of the forecast your business needed to act on first.

On the entry-level data specifically, the Stanford study and BLS projections both point at declining demand for routine, pattern-based accounting and bookkeeping work. Neither source we checked names accountants directly in its published examples. But the work described, transaction-level and repeatable, matches the category the AI 2027 scenario always flagged as the first wave. Treat the connection as a reasonable read of the data, not a direct quote from either study.

What should you check next quarter?

The advice from our original article holds under both branches: log agent decisions, keep more than one vendor and pay staff to verify AI output.

Here's the useful part. You don't need to bet on which track keeps moving faster. These three things were true before this update and stayed true after it.

Audit your agent's decision trail. If a QuickBooks or Sage agent categorizes a transaction, ask whether you can reconstruct why it made that call. Auditors are already asking for this under the current PCAOB posture on AI-assisted procedures. If your vendor can't answer clearly, note it before renewal.

Keep more than one vendor relationship. Frontier consolidation risk hasn't shown up yet, but concentration among accounting software vendors is happening anyway as agentic features become the selling point. Don't lock into an exclusive contract for a discount you'll regret when a competitor ships a better agent next year.

Pay your best junior to become your AI reviewer. The entry-level squeeze is real in the data. Someone who can catch an agent's mistakes is becoming the scarce, valuable skill on a finance team. A junior accountant who's good at spotting errors is the role to build around, not eliminate.

We'll keep scoring this forecast every quarter in the Nexairi Dispatch, our free newsletter for finance and accounting operators. If you want the next update before the trade press catches it, that's where it lands.

The forecast's biggest promises are running behind. Its quietest ones already reshaped how your books get closed. That asymmetry, not the headline predictions, is the actual news one year in.