AI Forecast Tracker
← Back to forecasts

Companies Are Betting Billions on AI Agents. Most Will Fail.

Will more than 60% of enterprise AI agent projects be abandoned or scaled back by end of 2027?

If your company is betting on AI agents, there's a 60% chance the project gets killed or downsized — but the 40% that succeed will be transformative.

Target: Dec 2027(572 days until resolution)
Assessed Probability
62%
Likely
Based on 4 expert predictions, 7 evidence items
Community Forecast
Cast your vote
Be the first to weigh in below

Your Prediction

Where do you think this lands?

Join others who've weighed in

5%95%
50% — More likely than not
MIT found that 95% of enterprise AI pilots deliver zero measurable ROI. S&P Global reports 42% of companies already scrapped most AI initiatives in 2025. But the narrative is shifting faster than expected. Enterprise agentic adoption hit 57% — blowing past Gartner's 40% prediction for end 2026. Salesforce Agentforce reached $800M ARR with 18,500 customers averaging 12 agents per organization. Claude agent teams are in production. The acceleration in framework quality since late 2025 is real. Most enterprise AI projects will still fail — that's the base rate for any enterprise technology (60-80%). But the success rate is improving faster than prior technology waves, and the successful deployments are delivering outsized returns. The AI-native startups (Harvey, Distyl, Basis) prove that the technology works — the enterprise adoption gap is about governance and integration, not capability.

Scenarios

Current value: 88-95% AI pilot failure rate (various surveys, 2025); but 57% enterprise agentic adoption (exceeding Gartner's 40% prediction)

S-curve position: Entering slope of enlightenment — early adopters proving value, late majority still struggling

Bear Case

75% failure (complexity overwhelming, runaway costs, but lower than before due to framework maturation)

Base Case

55-65% failure (improving from historical base rate, concentrated in companies without AI expertise)

Bull Case

40% failure (frameworks mature fast, governance commoditized, AI-native companies set the template)

How We'll Know

What we measure
Percentage of enterprise AI agent deployments that are abandoned, significantly scaled back, or fail to meet stated objectives per analyst surveys
Confirmed if
Analyst firms report >60% failure/abandonment rate for enterprise agentic AI projects by end 2027
Refuted if
Failure rate stays below 40%, OR successful agent deployments clearly outnumber failures
Data sources
  • Gartner Hype Cycle reports
  • McKinsey AI adoption surveys
  • S&P Global enterprise AI tracker
  • Industry earnings calls mentioning agent project outcomes

Evidence Trail

Evidence For

  • Mar 7, 2026

    MIT: 95% of AI pilots deliver zero ROI. S&P Global: 42% of companies scrapped most AI initiatives (up from 17%). Gartner: >40% of agentic AI projects expected to fail. 90% of orgs lack internal capability to scale agents. Historical enterprise AI project failure rate 60-80%.→ Probability: 70%

  • Mar 7, 2026

    Enterprise agentic adoption at 57% (exceeded Gartner's 40% end-2026 target). Salesforce Agentforce $800M ARR, 18,500 customers, avg 12 agents per org. Claude agent teams in production. AI-native startups (Harvey $195M ARR, Distyl $1.8B, Basis $1.15B) proving the technology works when built from scratch. Framework quality improving rapidly.→ Probability: 60%

  • Mar 9, 2026

    Goldman Sachs developing autonomous agents for trade accounting and onboarding. Lloyds Banking Group expects enterprise-wide agentic AI in 2026 projecting £100M added value. JPMorgan saving $1.5B through AI. Gartner projects 40% of enterprise apps embed agents by end 2026 (up from <5% in 2025). Systematic evaluation tools increase agent production success rate by 600%.→ Probability: 55%

  • May 16, 2026

    Gartner (May 5, 2026): approximately 80% of organizations are conducting AI-related workforce reductions, but those reductions are not delivering measurable ROI — autonomous-business cost savings being recycled into further AI investment rather than returning value. Intercom 2026 Customer Service Transformation Report: only 10% of teams have achieved mature AI deployment; 25-point performance gap from deployment depth, not decision. Writer survey 97% deployed but 29% see significant ROI. Grant Thornton: 78% of executives lack confidence to pass an independent AI governance audit within 90 days. Walmart CFO acknowledged AI 'has not impacted top-line sales yet'. Pattern: deployment without value capture is now multi-source-confirmed.→ Probability: 58%

  • May 23, 2026

    Medium (Snehal Singh) analysis of 847 enterprise AI agent deployments in 2026: 76% production failure rate; Forrester root-cause breakdown — 41% unclear success criteria, 33% insufficient tool/data access, 26% evaluation drift; HCLTech separately warned 43% of enterprise AI initiatives may fail. Gartner (May 5) found ~80% of organizations conducting AI workforce reductions are NOT seeing measurable ROI — autonomous-business cost savings being recycled into more AI investment without performance proof. Anthropic Economic Index (Jan 2026) showed augmentation up from 47% (Aug 2025) to 52% — partly product-driven, but slows the pure-automation case.→ Probability: 60%

  • May 30, 2026

    W22 — first peer-reviewed mechanistic quantification: MSR 2026 study (Mining Software Repositories conference, 11,771 GitHub PRs analyzed) found AI agents introduce 79.15% of CI failures while performing only 60.63% of corresponding fixes — a net negative failure balance even though AI fixes are 4x faster than human fixes (17.23 vs 71.70 minutes median). Separately, an enterprise agentic governance survey published this week found 72% of enterprises report production AI agents but 60% have NO governance framework for them — and 79% of organizations report adoption challenges, double the 2025 rate. Deloitte State of AI in Enterprise: 54% of C-suite admit AI adoption is "tearing their company apart". The 76% deployment-failure number from W21 now has an upstream mechanism: agents introduce more bugs than they fix.→ Probability: 62%

Evidence Against

  • Mar 7, 2026

    Salesforce Agentforce $800M ARR with 18,500 customers suggests real value delivery. 57% of enterprises running multi-step agent workflows. Agent frameworks improving rapidly (LangChain, CrewAI, Claude agent teams). Enterprise AI spend still growing.

How Our View Evolved

  • May 30, 202660%62%

    +0.02 → 0.62. MSR 2026 peer-reviewed study (11,771 GitHub PRs analyzed) puts the first mechanistic number on the failure side: AI agents introduce 79.15% of CI failures and fix only 60.63% of them — net negative even though AI fix-time is 4x faster than human (17.23 vs 71.70 min median). Enterprise governance survey: 72% report production AI agents but 60% have no governance framework; 79% report adoption challenges (double 2025). Deloitte State of AI in Enterprise 2026: 54% of C-suite say AI adoption is "tearing the company apart". W21 76% deployment-failure rate now has an upstream cause — agents introduce more bugs than they fix.

  • May 23, 202658%60%

    +0.02 → 0.60. Independent analysis of 847 enterprise AI agent deployments in 2026 reports 76% failure rate, with Forrester root-cause attributing 41% to unclear success criteria, 33% to insufficient tool/data access, 26% to evaluation drift — none are fundamentally model-quality failures. Gartner separately found ~80% of organizations conducting AI workforce reductions are seeing zero correlation with ROI; HCLTech warned 43% of enterprise AI initiatives may fail. Compound failure math is unforgiving: 85% per-step reliability gives ~20% success across a 10-step workflow.

  • May 16, 202655%58%

    +0.03 → 0.58. Gartner May 5 finding: ~80% of organizations conducting AI workforce reductions, but those reductions are NOT delivering measurable ROI. Intercom 2026 Customer Service report: only 10% of teams have achieved mature AI deployment, 25-point performance gap from deployment depth not deployment decision. Writer 97% deployed / 29% ROI gap, Grant Thornton 78% lack confidence to pass AI governance audit. Pattern: deployment without value capture is now multi-source-confirmed.

  • Mar 9, 202660%55%

    Financial services showing real agent value: Goldman/Lloyds/JPMorgan deploying agents with measurable ROI. Gartner projects 40% of enterprise apps embed agents by end 2026. High-end success stories temper the failure thesis.

  • Mar 8, 2026Initial assessment: 60%

    Baseline — initial published assessment

What Experts Say

Andrej Karpathy

AI Researcher, former Tesla AI Director, educator

Track record: 8/10
It will take 10 years to build all the agents that can do meaningful work
Jan 2026 | interview
We assess this claim as 70% likely

Gartner

Technology Research and Advisory

Track record: 7/10
40% of enterprise applications will feature task-specific AI agents by end of 2026, up from <5% in 2025
Aug 2025 | report
We assess this claim as 91% near certain

Gartner

Technology Research and Advisory

Track record: 7/10
Over 40% of agentic AI projects will fail by 2027 due to unintended decisions, runaway costs, and governance challenges
Oct 2025 | report
We assess this claim as 81% very likely

Gartner

Technology Research and Advisory

Track record: 7/10
Agentic AI could drive ~30% of enterprise application software revenue (~$450B) by 2035
Aug 2025 | report
We assess this claim as 30% unlikely

What Could Go Wrong

Agent tooling matures faster than expected. A killer framework emerges that makes agent governance easy. 'Failure' definition is too broad — many scaled-back projects still deliver partial value. The 60% threshold is actually the historical base rate, making this almost a tautology.

What should we track about this topic?