Zan Digital › Companies
MIT
Every piece of Zan Digital analysis that covers MIT — 26 pieces, each figure traced to a named source.
- 001AI Capex Is in the GDP Data. The Payoff Is NotAI investment drove 39% of US GDP growth in early 2025. The productivity return has not arrived. Whether that is a lag or an absence, and how to tell.39%
- 002AI Spending vs EBIT: Gartner's $2.7T Meets McKinsey's 37%Gartner puts 2026 AI spending at $2.7T, up 49.5%. McKinsey says firms reporting any EBIT impact hold flat at 37%. Both methods, side by side, and the fix.$2.67T
- 003Dot-Com vs AI Bubble: Six Metrics Where It Breaks DownEveryone compares 2026 to 2000. On six metrics, from P/E ratios to profitability, the comparison mostly fails. Here is the one place it does hold.60x vs 23x
- 004GenAI ROI: Why 95% of Pilots Show No Return, and Who DoesMIT found 95% of GenAI pilots show no measurable P&L return. What the 5% did differently, where the money is really being made, and how to measure your own.95%
- 005MIT Iceberg Index: What the 11.7% Figure MeasuresMIT's Iceberg Index puts 11.7% of US wage value within reach of current AI. It is a capability map, not a job-loss count. How it works and where it breaks.11.7%
- 006The IMF Warned on AI Valuations. The Optimists AnsweredThe IMF flagged correction risk in AI-exporting economies. Supporters argue compute scarcity makes the spending rational. Both cases, and what would settle it.Jul 2026
- 007AI Moats in 2026: Integration Depth Beats Model DepthThe top 4 AI labs sit within 25 Elo points of each other. Model quality no longer separates products. Integration depth does, and here is the evidence.25 Elo
- 008Seat Compression: When Success Destroys Your RevenuePer-seat pricing fell to roughly 15% of the SaaS market. If your product reduces headcount, it reduces your revenue. The trap, and the four ways out.~15%
- 009AI Code Review Tools Compared: Price and ProofAI code review tools cost $24 to $72 per developer on CodeRabbit, $40 on Graphite, and metered credits on Copilot. See triggers, platforms and the evidence.$24 to $72
- 010AI Gateway Comparison: Fees, BYOK and Failover TermsOpenRouter takes 5.5% on credits, Cloudflare 5% on Unified Billing, Vercel 0% on tokens. Four AI gateways compared on fees, BYOK, retention and failover.5.5%
- 011LLM Testing Tools: Promptfoo vs DeepEval vs RagasPromptfoo lists 44 non-LLM assertions, DeepEval passes at 0.5 by default, Ragas has no CI gate. LLM testing tools compared on one 12-case agent suite.44
- 012Open Weights in Production: Qwen, DeepSeek, Llama, GLMChinese models take 61% of OpenRouter tokens and about 1% of enterprise LLM spend. Comparing Qwen, DeepSeek, Llama and GLM on quality, cost and licence.61%
- 013OpenAI Evals Shutdown: 4 Migration Paths ComparedOpenAI Evals went read-only on 31 October and shuts on 30 November 2026. We compare 4 destinations on dataset import, graders and price, plus a checklist.30 Nov
- 014AI Agent Adoption Surveys: Why 70% and 97% Don't Add UpCreatio says 70% are past AI experimentation. WRITER says 97% of executives deployed agents. Different samples, sponsors and definitions, side by side.70%
- 015AI Incident Database vs OECD: 362 a Year, 324 a MonthThe AI Incident Database logged 362 AI incidents in 2025. The OECD monitor logged about 324 a month. Inclusion rules, not AI reliability, drive the gap.362
- 016AI Proof of Concept to Production: 8 Tests to Run FirstDeloitte reports 38% of organisations piloting agentic AI and 11% running it in production. Here are 8 production-shaped tests that close that gap.11%
- 017Federal AI Use Case Inventory: 1,818 of 3,611 in UseThe 2025 federal AI use case inventory lists 3,611 uses at 56 agencies. 1,818 are piloted or deployed, and between 878 and 1,410 are in production.3,611
- 018MIT 95% AI Pilot Failure Study: What 52 Interviews ShowMIT NANDA's 95% figure rests on 52 interviews and 153 conference surveys. Of firms that piloted custom AI tools, 25% reached production. The method, read.52
- 019Scaling AI Agents: What the 25% Did DifferentlyOnly 25% of organisations moved 40% or more of their AI pilots into production. Deloitte, McKinsey and BCG agree on what that minority did differently.25%
- 020Starbucks AI Inventory: 11,000 Stores, Scrapped in 9 MonthsStarbucks dropped its AI inventory counter from 11,000+ stores 212 days after launch. A ledger of 7 retail AI rollbacks sets stated reasons against the record.11,000+
- 021Open Source AI Licence: The 4 Clauses Nobody ReadsMost open weight models are not open source. 4 clause types decide commercial use, from Llama's 700 million user cap to terms you must pass downstream.700M
- 022AI Upskilling Pledges: 23M People Promised, No Job DataAmazon, Microsoft and Google pledged AI training to 23 million people. We set each target against reported progress, the definition of trained and job data.23M
- 023Measuring Developer Productivity With AI: The 19% ProblemMETR found AI made experienced developers 19% slower while they felt 20% faster. What DORA and SPACE measure instead, and a scorecard you can defend.19%
- 024Product Manager Skills When AI Writes 75% of CodeAI now writes 75% of new code at Google. That moves the product manager job upstream, toward specification quality, evaluation design and problem selection.75%
- 025Spec Writing for AI Agents: Spec Kit vs Kiro ArtifactsSpec Kit asks for 7 artifacts per feature, Kiro asks for 3. What each makes engineers write, a template that fits both, and who should own spec writing.7 vs 3
- 026Stanford AI Jobs Study vs Yale: 13% Drop, No DisruptionStanford's ADP payroll study found a 13% employment drop for 22 to 25 year olds in AI-exposed jobs. Yale's survey found no disruption. Why both can hold.13%