Tracking a Forecast Against Reality
The AI 2027 Ledger
In April 2026 we translated the AI 2027 forecast into what it means for business owners, and promised a quarterly scorecard: did each prediction land, slip, or miss? This is that scorecard, updated as each checkpoint arrives.
Entries
7
Confirmed
3
Updated
Apr 27, 2026
Latest Entry
March 2027: Agent-3 and superhuman coding
A model matching or exceeding top human coders, with acute disruption predicted first for junior software engineering roles and comparable pressure expected on junior bookkeeping and accounting roles.
| Checkpoint | Prediction | Status | Source |
|---|---|---|---|
| March 2027: Agent-3 and superhuman coding | A model matching or exceeding top human coders, with acute disruption predicted first for junior software engineering roles and comparable pressure expected on junior bookkeeping and accounting roles. | Pending - monitor hiring signals Q1 2027 | AI 2027 Scenario Forecast |
| February 2027: Model weight theft | A predicted breach in which Agent-2's model weights are stolen, with the downstream business risk being accelerated vendor consolidation and higher switching costs for SMBs. | Pending - geopolitical trigger not yet observed | AI 2027 Scenario Forecast |
| January 2027: Agent-2 and neuralese | Internal reasoning that becomes opaque to human monitors, raising a direct audit controls problem: no visibility into how an AI agent reached a decision affecting financial records. | Pending - watch Q1 2027 lab releases | AI 2027 Scenario Forecast |
| Late 2026: Agent-1-mini | A cheaper, fine-tunable model economical to embed directly into consumer and SMB software, with vendor roadmaps pointing to embedded agents in accounting, CRM, and automation tools, alongside projected 12-25% SaaS renewal increases. | In progress - pilot availability emerging as of April 2026 | AI 2027 Scenario Forecast |
| Early 2026: 1.5x R&D multiplier | A predicted AI research speed multiplier, visible on the business side as faster SaaS feature releases across ChatGPT, Claude, and Copilot. | Confirmed | AI 2027 Scenario Forecast |
| Late 2025: Agent-1 trained | A model trained with roughly 1,000x the compute spent on GPT-4, optimized for AI research acceleration. Publicly verifiable training details are limited, but labs' public emphasis on agents and AI-assisted research tracks the direction. | Directionally confirmed | AI 2027 Scenario Forecast |
| Mid-2025: First agents emerge | AI 2027 predicted "stumbling agents" capable of simple tasks like ordering burritos and summing expenses, with bookkeeping named as the canonical first-wave task. | Confirmed | AI 2027 Scenario Forecast |
Every row links to its source. If a status looks wrong or you have a correction with a source attached, tell us — this ledger is only useful if it stays accurate.
Join Nexairi Dispatch
AI is showing up everywhere you live and work. A short dispatch, 3x per week, so you see it coming.