Writing

Frameworks and honest lessons from enterprise AI.


Essays

Longer arguments and thesis pieces.

Posts

Shorter pieces that started on LinkedIn.

Most AI Agent Failures Aren't Alignment Failures. They're Architecture Failures.

A new paper reframes hallucination and overconfidence as symptoms of unbounded autonomy — and proposes a four-state framework (SMARt) for governing when agents can and cannot act.

4 min · June 2026

Specialized AI Agents Can't Do Spreadsheets. General Agents Can.

WorkstreamBench tested agents on real financial modeling tasks. Claude's general web interface scored 69.1. Its specialized Excel add-in scored 60.4. Same model, different harness, 3.8× gap.

3 min · May 2026

A 3B Model Just Outperformed GPT-4o by Internalizing Skills Into Its Weights

SKILL0 trains small models with curated skills then progressively withdraws them. The result: a 3B model scores 87.9% where GPT-4o scores 48.0%, at 5× lower inference cost.

4 min · May 2026

OpenAI Just Created a $4B Company to Solve a Problem Most Enterprises Still Think Is a Model Problem

The OpenAI Deployment Company launched with $4B, 19 investors including McKinsey and Bain, and 150 engineers from Tomoro. The model provider just ate the consulting layer.

4 min · May 2026

90% of Firms Report Zero AI Productivity Impact. Yet 80% Already Laid Off Workers.

NBER data shows 90% of firms saw no measurable productivity change from AI. Gartner shows zero correlation between AI layoffs and AI ROI. The layoff narrative is not connected to the results.

3 min · May 2026

SAP Just Announced 'Autonomous Enterprise.' 50 Years of ERP Rewritten with AI Agents.

SAP Sapphire 2026: 200+ AI agents, Anthropic Claude powering HR/procurement/supply chain, and a bet that governance — not models — is the winning layer.

4 min · May 2026

A Coding Agent Just Scored the Theoretical Maximum on Atari Breakout. No Neural Network. No Gradients.

Heuristic Learning uses coding agents to maintain and improve software-based policies through feedback loops — matching Deep RL baselines without training a single neural network.

4 min · May 2026

Your AI Costs Might Be 30-40% Higher Than They Need to Be — Because of the Tokenizer

1,308 models trained to show that the Chinchilla rule of ~20 tokens per parameter is wrong. The universal rule is ~60 bytes per parameter.

4 min · May 2026

An 8B Model Just Outperformed a 235B Model on Agent Tool-Use. The Difference Was Environments, Not Parameters.

Agent-World builds 2,000 realistic training environments with 19,822 tools — and shows environment diversity beats parameter scaling for agentic tasks.

4 min · May 2026

64% of Enterprises Deployed AI Agents Before They Were Ready. They're Now Paying for It.

Monte Carlo's survey of 260 enterprise leaders reveals: 63% found unauthorized data access, 36% can't roll back a failing agent, 70% expect to rebuild systems they already shipped.

3 min · May 2026

Google Renamed Vertex AI. I Was There When the Original Strategy Was Shaped.

Google Cloud Next 2026: Vertex AI becomes Gemini Enterprise Agent Platform. A2A protocol at 150 orgs in production. 200+ models including Anthropic Claude. A Google insider's perspective.

4 min · April 2026

Memory, Skills, and Rules Aren't Three Different Systems. They're the Same System at Different Compression Levels.

A new paper unifies agent memory, skill discovery, and rule learning into a single Experience Compression Spectrum — and shows agents with Level 2 skills outperform Level 1 memory by 68.5 percentage points.

4 min · April 2026

DeepSeek V4: 1M Context at 21× Less Than Opus. The Open-Weight Gap Just Closed Again.

DeepSeek shipped V4 the same day as GPT-5.5. 1.6T parameters, 1M context, Apache 2.0, $3.48 per million output tokens. Sparse attention cuts KV cache to 10% of V3.2.

4 min · April 2026

22 AI Agent Frameworks Tested Head-to-Head. The Top 12 Were Separated by 1.4 Percentage Points.

The largest systematic comparison of agentic AI frameworks reveals that engineering discipline — memory management, retry policies, failure handling — matters more than architecture.

4 min · April 2026

Why Your AI Costs Go Up With Usage and Theirs Go Down

A new paper shows 84.6% token savings through knowledge compounding. The insight: tokens should be capital investments, not consumables.

4 min · April 2026

Most People Think Improving an AI Agent Means Retraining the Model. That's One Lever. There Are Four.

A new paper shows agents improving themselves without model updates. But the bigger insight is the full optimization surface most enterprises aren't using.

4 min · April 2026

Are You Running 5 Agents Because the Task Needs 5?

A new paper tested when to collapse multi-agent systems into a single agent. The answer is counterintuitive: it's not about the task. It's about the metric.

4 min · April 2026

Your AI Agents Are Acting As You. Literally.

When your Copilot agent sends an email, it does it under your identity. Your credentials. Your permissions. Microsoft's Agent 365 won't fix this until December 2026.

4 min · April 2026

Jensen Says AGI. The Benchmark Says 0.26%.

Jensen Huang told Lex Fridman he thinks we've achieved AGI. The same week, ARC-AGI-3 launched. Humans score 100%. Best frontier AI scored 0.26%. Both are right.

3 min · March 2026

An AI Agent Just Beat NVIDIA's Own Engineers at Optimizing NVIDIA's Own GPU

The AVO paper gave a coding agent access to CUDA docs and Blackwell B200 specs. It produced 40 kernel versions in 7 days — outperforming cuDNN by 3.5% and FlashAttention-4 by 10.5%.

4 min · March 2026

You're Not the Engineer Anymore. You're the Director.

Andrej Karpathy hasn't typed a line of code since December. He spends his days expressing his will to agents. The bottleneck just changed.

3 min · March 2026

We're Hiring a Staff AI Scientist at Uniphore

This is the role I wish existed when I was still at Google: applied research close enough to production that the work actually ships.

2 min · March 2026

Jensen Compared OpenClaw to Linux. Nobody Is Talking About What That Means.

If the agent runtime is Linux, then the runtime is a commodity. Every enterprise will run the same one. The value migrates up — to the intelligence layer.

3 min · March 2026

Cursor Got Caught. I'm Reading It as a Strategy Guide.

Their new 'proprietary' model was Kimi K2.5, an open-weight model from Moonshot AI, fine-tuned with RL. A developer found it in 24 hours. The internet called it a scandal.

2 min · March 2026

Your AI Bill Is Growing. Your AI Moat Isn't.

Midjourney runs on models they own. Their inference costs drop every quarter. Your enterprise runs on Claude and GPT. Your costs go up with usage. Your moat is zero.

4 min · March 2026

Knuth Said 'Shock! Shock!' — And If You Know Knuth, You Know What That Means

Donald Knuth, the most respected computer scientist alive, was stunned when Claude solved an open problem he'd been working on for weeks. The implications go beyond math.

3 min · March 2026

The Most Unfair Advantage in the History of Work

Andy Grove wrote the most important sentence in management. In 2026, that equation quietly changed. Your organization now includes AI agents.

3 min · March 2026

MCP vs CLI Is the Wrong Debate

We just gave an AI agent unrestricted CLI access to our staging environment. The security team found out from a Slack alert — not a governance review.

3 min · March 2026

Most Enterprises Are Stuck at Stage 2

You have deployed AI. It works. But it does not compound. That is not Stage 3. That is Stage 2 — and most enterprises are stuck there.

4 min · March 2026

AI Agents Are About to Cost More Than Offshore Labor

The technology we built to replace offshore labor is about to cost more than offshore labor. A Gartner prediction that should worry every enterprise AI buyer.

3 min · March 2026

Vibe Coding Will Kill the Agent Builder UI

In 2024, we built drag-and-drop agent builders. In 2026, developers just describe what they want. The config UI era lasted 18 months.

4 min · March 2026

One Lesson About Leadership

One lesson about leadership has stayed with me throughout my career. Leadership is not about doing one thing. It is about making one team.

3 min · February 2026

Total Anthropic Victory (For Now)

Everything they have done in 2026: $20B revenue run rate, number one on iOS, the Pentagon. A breakdown of how Anthropic is winning — and the risks ahead.

5 min · February 2026

I publish on LinkedIn and update this site as I build.