All articles
Every article published on A-Eye Level, collected here.
-
It Checked Its Own Work
Anthropic's Claude caught its own analysis error in a lab test, then an outside lab confirmed it was right. Here's why that matters more than speed.
7 min read -
Four Levers, Not One Dial
AI autonomy isn't one dial. Anthropic's own team breaks it into four separate handoffs, and only one of them decides whether the rest are safe to hand off.
6 min read -
The Silent Exit Clause
A safety check the builder can veto isn't a check. Anthropic's own policy confirms the gap, and METR's audit shows what real independence looks like.
4 min read -
The Check That Left With the Afternoon
A McKinsey case describes a distributor whose system reads public building permits and drafts the sales outreach. The work it removed was somebody's afternoon, and an afternoon is never only one job: while a person was assembling the material, they were also reading it. That reading was a control nobody had written down, and it leaves with the work.
4 min read -
Don't Build What You Can't Keep
McKinsey's 2026 survey found 32% of respondents saying their organization decided against buying at least one software product or feature because it could be built with coding agents. The companion post argued that the buy rule grew a second half. This article does the three things the post could not: reads the survey's size gap honestly, sorts builds into the three kinds that need different rules, and shows how to price what keeping a built tool costs every month.
9 min read -
Cheap to Write, Expensive to Keep
A system that remembers across more than one layer has a stratum you can delete with no consequence and a stratum that takes everything with it. Cost asymmetry moves important knowledge into the cheap one without anyone deciding it. This is the evidence for why that happens, and why better writing tools do not fix it.
7 min read -
What Stays Expensive
The peer-reviewed version of a landmark AI productivity study found the gains landed on the least skilled and quality slipped slightly for the most skilled. That is not a productivity result. It is a repricing of what you have been paying for.
8 min read -
Assessment or Echo
An agent's recommendation is assembled out of what it wrote down and how you asked. Neither of those is visible in the recommendation itself.
7 min read -
An Agent Is Not an App
Software waits. An agent acts. That single difference is why agent projects run on a hiring clock, and why the invoice is the wrong document to plan from.
8 min read -
A Retention Policy for AI Meeting Agents
Connect an agent to your meetings and you set a retention policy by default. Here it is written down: five artifacts, three gates, four exceptions.
8 min read -
The Verification Step Nobody Owns
Deloitte, KPMG and EY each shipped a report with fabricated citations in eight months. The failure was economic: verification nobody priced.
9 min read -
Redesign the Work, Not the Tool
Deloitte found 66% of leaders say human-AI work design is critical, but only 6% do it. The gap is work design, not technology, and here is the fix.
6 min read -
The Regulator Wants to Be a Shareholder
OpenAI is reportedly discussing giving the US government a 5% stake. As the state shifts from rule-setter to shareholder, who oversees whom?
6 min read -
The Road to Cooling a City Starts From the Roof Up
Google turns satellite imagery into a building-by-building map of which roofs to cool first. The harder question is who chooses the priority.
6 min read -
Compliance Was the Floor. Now Set Your Own Bar.
In one month the EU, Colorado, and the White House eased AI rules. Compliance was a floor others set. The bar is yours now, and five moves to set it.
9 min read -
Tables, Not Language
SAP bet over 1 billion euros on AI that does not chat. Why tabular foundation models matter for CEOs, and the first move to make.
5 min read -
Who Owns the Learning Loop?
Nadella named the asset: token capital. Owning it means keeping your data and definitions portable while you rent the model. Here is the three-layer test.
6 min read -
Why Nothing Changed After Your Team Adopted AI
Logins are up, the work is identical. The AI adoption-to-transformation gap is a measurement problem leadership owns. Four moves to fix the scorecard.
7 min read -
Your AI Adoption Dashboard Is Not a Strategy
Adoption metrics report this quarter. The two legs that decide your AI strategy do not. The leading-vs-lagging trap, and a one-hour test to run on Monday.
7 min read -
Stop Importing Your AI Workforce Forecast
AI founders keep reversing their jobs forecasts. Your real workforce signal is three internal numbers that move before the macro does.
6 min read -
Already On Your Payroll: The Agent Manager Audit
The agent manager is already on your payroll. A four-pillar audit covering succession risk, performance, comp, and the board agenda CEOs don't have.
8 min read -
Verifier vs Self-Report: Where Your AI Governance Lives
Your AI is grading its own homework. Three operator tests and a five-question board audit to find where that gap lives in your workflow.
7 min read -
The CAIO Confession: What the Seat Actually Owns
Three out of four organizations now have a Chief AI Officer. Two precedents, three operator tests, and a five-question board audit before you hire.
7 min read -
Is Your AI Strategy Theater? A Three-Test Boardroom Audit
75% of executives say their AI strategy is 'more for show.' Three operator tests separate the 25% with real strategy from the rest.
8 min read -
Where the Harness Pays Off, and Where It Does Not
Most AI budgets fund the wrong remedy because the workflow type was never diagnosed. Capability-bound vs friction-bound is the missing one-page policy.
9 min read -
The Boardroom Risk of Confident AI
Most AI policies don't yet name confidence as a governance surface. The four-item lens is the smallest piece of policy that closes the gap.
8 min read -
What AI-Native Teams Lose When They Win
Coinbase made lean-AI-native official. The CEO question isn't whether AI flattens the org. It is what gets removed when manager layers go.
4 min read -
Your Agents Already Think Shutdown Is Optional
Palisade documented agents resisting shutdown. Kiro, Meta, and a $437 LangChain runaway proved it in production. Three CEO questions to ask now.
3 min read -
When AI Capability Tier Becomes a Balance-Sheet Line
Anthropic's Project Deal showed better AI agents capture $3.64 more per transaction. The cohort running the cheaper one never noticed.
5 min read -
The Founders Who Win Don't Delegate Understanding
Walton, Schultz, Nadella, Chesky, and Schreiber show the pattern behind every paradigm shift, and what it means for AI in your company today.
10 min read -
AI Is a Mirror: Stanford's Lesson From 51 Deployments
Stanford studied 51 successful enterprise AI deployments. The difference was never the AI model. It was the organization. Here is what that means.
8 min read -
When to Override an AI Agent: The Three-Axis Threshold
Only one in five companies have mature AI agent governance. Three axes for when to override the agent's call: dollars, reversibility, audit-trail depth.
3 min read -
Why 20% of Companies Capture 74% of AI Returns
PwC: 20% of 1,217 companies capture 74% of AI returns at 7.2x performance. Winners invest 2.5x more AND redesign around AI. Inside: the John Deere case.
3 min read -
Shadow AI: Five Questions the Board Should Ask the CISO
Cybernews: 93% of execs use shadow AI vs 59% of employees. Five questions the board should ask the CISO about the two-tier audit surface.
4 min read -
Four Allocation Rules for the 2026 AI Budget
Stanford's 2026 AI Index shows a 3-4x productivity spread across functions. Four allocation rules for the 2026 AI budget under the aggregation hedge.
4 min read -
Four Clauses Every AI Contract Should Carry in 2026
The 2026 AI Index shows enterprise governance advancing while vendor disclosure retreated. Four contract clauses close the gap at procurement.
5 min read -
Skills Beat Agents: The AI Layer CEOs Keep Missing
Agent count is the visible metric. Skill depth is the layer that actually produces returns, survives vendor changes, and becomes IP the company owns.
5 min read -
92% Spending More on AI. 56% Getting Nothing Back.
92% plan to increase AI spend. 56% see zero returns. The 6% pulling ahead redesigned workflows, not tool stacks.
4 min read -
The CEO's First AI Delegation Is Almost Always the Wrong One
Most CEOs delegate the wrong AI task first. A framework for picking the right starting point, and what never to delegate.
4 min read -
88% Resolved, 22% Loyal: The AI Customer Trust Gap
AI resolves 88% of customer issues, but only 22% prefer the company. Four studies reveal why efficiency metrics miss what drives loyalty.
4 min read -
Fewer AI Tools, Double the ROI: What 1,800 Executives Reveal
BCG surveyed 1,800 executives: companies focusing on 3.5 AI use cases generate 2.1x the ROI. Five Tier 1 sources confirm fewer AI bets win.
5 min read -
The Strongest Predictor of AI Adoption Isn't Technology
A Harvard-led NBER study found management encouragement, not technology, is the strongest predictor of AI adoption and the productivity gap is widening.
6 min read -
Meta Ranks Employees by AI Usage. History Says That Backfires.
Meta, OpenAI, and Shopify now rank employees by AI token consumption. A 1975 management paper explains why measuring usage volume backfires.
5 min read -
515 Startups Got Identical AI Tools. One Group Made 1.9x More.
515 startups got identical AI tools. The ones who saw how others reorganized around AI generated 1.9x more revenue. The bottleneck is not the technology.
6 min read -
70% of the S&P 500 Talk AI. Only 1% Measured What It Did.
Goldman Sachs found no economy-wide AI productivity link, but 30% gains in teams that measured specific use cases. The gap is measurement.
5 min read -
Oracle and Meta Cut Tens of Thousands of Jobs. It's Not AI.
Oracle is cutting 30,000 jobs on $124 billion in debt. Meta trims teams while doubling AI spending to $135 billion. The real reason is not AI efficiency.
4 min read -
How to Build an AI Model Strategy That Survives Market Shifts
Anthropic went from 12% to 40% of enterprise AI. OpenAI dropped from 50% to 27%. Here's how to pick models without getting locked in.
5 min read -
502,000 AI Layoffs Planned. Zero Productivity Evidence.
44% of U.S. CFOs plan AI-related layoffs this year, roughly 502,000 positions. Goldman Sachs found zero measurable link between AI and productivity.
4 min read -
What Zuckerberg's AI Deputy Reveals About Your Information Gap
Zuckerberg is building an AI agent to bypass the layers between him and his data. The real story is the information gap every CEO shares.
4 min read -
McKinsey Runs 25,000 AI Agents Next to 40,000 Humans
McKinsey runs 25,000 AI agents alongside 40,000 consultants. Their org structure offers a template most companies haven't started building.
4 min read -
The AI Commerce Shift: When Your Platform Vendor Decides for You
Shopify activated AI storefronts for millions of merchants without asking. Amazon and Google are next. Here is how to audit your own vendor stack.
4 min read -
The AI Talent Exodus: When the People Behind Your Tools Go
Mira Murati left OpenAI, raised $2 billion before shipping a product. Meta offered $1 billion to poach one engineer. Capital follows people, not products.
5 min read -
Google's TurboQuant: When Software Rewrites the AI Cost Equation
Google showed AI models can run on one-sixth the memory they need. Independent engineers confirmed it. Here is what it means for your AI budget.
5 min read -
OpenAI Killed Its Most Popular Product. The Math Behind That Decision.
OpenAI killed Sora six months after launch, 9.6 million downloads, a $1 billion Disney deal. Every GPU on video was a GPU not running ChatGPT.
4 min read -
The $700 Billion Foundation Under Your AI Strategy
The four largest cloud providers are investing $700 billion in AI infrastructure in 2026. For every company on the cloud, these numbers set the floor.
4 min read -
The Physical World Underneath AI
$1.5 trillion in AI capital assumes four physical things keep working: energy, chips, submarine cables, raw materials. Three are under stress now.
5 min read -
Your Best Managers Probably Feel Deep Empathy. Their Teams Can't Tell.
A study of 968 people found no link between feeling empathy and showing it. One AI coaching session closed the gap by nearly a full standard deviation.
4 min read -
66% of CEOs Froze Hiring for AI. The Real Problem Isn't Spending.
Corporate America cut 1.17 million jobs to fund AI. Only 20% of organizations redesigned how work gets done. That gap is where AI budgets die.
4 min read -
What 80,508 People Actually Think About AI
Anthropic surveyed 80,508 people across 159 countries. The average AI user holds 2.3 fears about the same technology they say is working.
5 min read -
$14 Billion in 60 Days: The Physical AI Signal
$14 billion in robotics and physical AI venture capital in 60 days, matching all of 2025. The biggest raises are not hardware leaders. They are data companies.
5 min read -
Your AI Tools Are Multiplying. Your People Aren't Keeping Up
BCG found the tipping point: at 3+ AI tools, productivity collapses. The workers hit hardest aren't the disengaged ones. They're the high performers.
4 min read -
Your Board Is Asking About AI. What Does Your Report Actually Say?
56% of CEOs report no financial benefit from AI. The problem is not adoption. Nobody built a framework for reporting what AI actually produced.
4 min read -
You're Not Using AI Wrong. You're Building Wrong.
McKinsey tested 25 factors. The single biggest predictor of AI profitability was workflow redesign. Not better models, not bigger budgets.
5 min read -
The AI Agent Governance Gap
81% of companies plan to expand AI agents this year. Their scaling plans almost never mention governance. Here's the gap that will cost them.
5 min read -
The Case for Using AI Less
More AI is not always better AI. When cognitive dependency replaces cognitive effort, the advantage reverses.
3 min read
