Track record
Dated, public predictions about where AI agent adoption breaks. Checked quarterly. Misses included.
Buyers increasingly refuse to pay six figures for a readiness assessment that a fixed-fee diagnostic answers, forcing advisory firms to unbundle discovery from delivery.
Buyers burned by unused seats push vendors to consumption or outcome pricing; new contracts shift materially away from flat per-seat.
Analyst questions shift from how much is being invested to what returned, and vague answers get punished.
Post-mortems on stalled programmes will name data access, quality and lineage more often than any model or vendor limitation.
A publicly disclosed incident will trace to an agent taking an unreviewed action with production credentials.
Organisations buying assistant seats org-wide will see under 40% weekly active usage twelve months post-purchase, making seat-based AI the largest single line of new shelfware.
After two years of departmental experimentation, mid-market organisations consolidate AI spend under one owner because nobody can answer what the total spend bought.
More than 40% of agentic AI projects started in 2025-2026 will be scrapped before reaching production, driven by unclear business value and escalating cost rather than model capability.
The majority of vendors currently marketing autonomous agents are wrapping existing model APIs. Consolidation and shutdowns will remove most of them from mid-market shortlists within two years.
Pilot-to-production conversion stays under 33% through 2026, with data readiness and change management — not model quality — as the binding constraints.
Three questions, an instant estimate of whether your pilot reaches production.
Check your own project's oddsWe review submissions weekly. Nothing publishes automatically — we remain the sole publisher of record for the board.
No submissions naming private individuals. Predictions about named public companies are reviewed before publishing.