Writing

Thinking out loud.

Perspectives on AI, engineering, and what it takes to build things that last.

Agent Engineering·May 20, 2026·25 min

When consequence becomes real

A Tuesday in late autumn, a new customer with three people inside their account, and the moment the agent's tool call could have run twice. The roles table, idempotency enforcement, and the outbox, what each one prevents and where each one earns its keep.

Agent Engineering·May 20, 2026

When the contract is downstream of your data

A Monday morning in early winter, an email from a fintech CISO, and the four million row events table that needs a hash chain by Thursday. The hash chain, the append only events, the actor_kind enum conversion, and the secret dereference layer that takes a week.

26 min

Agent Engineering·May 20, 2026

When measured pressure means something specific

A 3am pager from Singapore, a 200ms SLA the architecture cannot meet from one region, and the discipline of refusing to architect against pressure that has not been measured. Read replicas, multi region, sharding, semantic caching, and the meters that earn each decision.

25 min

Agent Engineering·May 18, 2026

When a second operator changes everything

A single Thursday afternoon scene before a design partner logs in Friday. One migration, three policy turn ons, and the eight check rhythm that buys a night of sleep.

24 min

Agent Engineering·May 15, 2026

What to build before you have a customer

Six Stage 1 decisions that cost almost nothing today and remove the Stage 2 rewrite. Plus the ten things to leave on the floor until pressure shows up.

26 min

Agent Engineering·May 12, 2026

When your vector database stops being enough

You shipped RAG. It solved a real problem. The next four failure modes are not retrieval problems. Here is where they actually live.

15 min

AI Strategy·Apr 29, 2026

Who signs off when your AI agent spends $42,000?

Your AI agent is about to approve a $42,000 refund. Who signs? Here is the one boundary every financial action needs before money moves.

7 min read

Compliance·Apr 26, 2026

The EU AI Act Is 98 Days Away. What Your Company Needs to Do Now.

Article 12 of the EU AI Act binds 2026-08-02. A plain-English guide for compliance officers, legal leads, and heads of ops. No jargon.

8 min read

Engineering·Apr 23, 2026

The v0.7 Bet: Postgres Stays Authoritative, the Bridge Fails Open

We shipped v0.7.2 of the Code Atelier Governance SDK. Here is how an opt-in dual-write bridge gives operators a hosted audit dashboard without new infra.

11 min read

AI & Commerce·Apr 20, 2026

Your Next Biggest Customer Will Not Be Human - Here Is How to Win Their Business

Gartner projects that AI agents will intermediate over $15 trillion in B2B purchases by 2028. Google, Shopify, Stripe, and Visa are building the infrastructure right now. For CEOs whose revenue depends on someone choosing to buy from them, this is the most significant new sales channel since the internet itself.

11 min read

Automation·Apr 17, 2026

The 90-Day AI Automation Playbook: How Operations Teams Are Reclaiming 40% of Their Week

McKinsey estimates 57% of U.S. work hours are now technically automatable. Yet the companies seeing real results are not automating everything at once. They follow a proven 90-day sequence: audit, build, scale. Here is the playbook that turns 200 hours of weekly manual work into 80+ hours of reclaimed capacity.

15 min read

Cybersecurity·Apr 14, 2026

What the LiteLLM Supply Chain Attack Teaches Us About Securing AI Infrastructure

On March 24, 2026, attackers published backdoored versions of LiteLLM to PyPI using credentials stolen through a compromised security scanner in LiteLLM's own CI/CD pipeline. The incident exposed a pattern we see across nearly every AI deployment: the packages that aggregate the most credentials receive the least scrutiny. Organizations that treated their AI supply chain like critical infrastructure were unaffected. Here is what they did differently.

11 min read

Automation·Apr 10, 2026

5 AI Automations That Pay for Themselves in 30 Days

Your team is losing 30 to 50 hours per week to tasks that AI can handle today. Here are five automations - from email triage to automated reporting - with specific cost breakdowns, payback timelines, and real results from teams that made the switch.

14 min read

AI Strategy·Apr 9, 2026

The Agentic Risk Standard Does Not Exist Yet - Here Is What Executives Are Evaluating Instead

Buyers are approving agent deployments without a shared risk standard. Here are the ten dimensions a mature one would have to cover, and how to ask today.

12 min read

Engineering Leadership·Apr 9, 2026

When Your Startup Needs a CTO But Cannot Afford One - The Fractional Leadership Model That Works

Most Series A and B startups hit a point where they are making critical technology decisions without senior technical leadership. A fractional CTO - a seasoned executive embedded in your team 2-3 days per week - can close that gap at a fraction of the cost. Here is how to know when you need one, what they actually do, and why the model works.

14 min read

AI Strategy·Apr 7, 2026

Microsoft Just Launched Its Own AI Models - Why Your Vendor Lock-In Strategy Needs an Update

On April 2, Microsoft released MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 through Microsoft Foundry - competing directly with OpenAI and Google. With the Microsoft-OpenAI partnership visibly fracturing, enterprise leaders need a vendor strategy that builds optionality rather than dependency. Here is the decision framework.

13 min read

Security·Apr 5, 2026

The Agent IAM Playbook: How Enterprises Are Securing Their Non-Human Workforce

With Okta launching its AI Agents platform on April 30 and Microsoft confirming that agents are now both workforce and attack surface, enterprise identity management has reached an inflection point. Here is a practical maturity model for securing AI agent identities - and the three steps experienced teams implement first.

11 min read

AI Strategy·Apr 4, 2026

How Companies Are Cutting AI Costs 60% Without Cutting AI Investment

Per-token AI costs dropped 280x in two years, yet enterprise AI bills jumped 320%. The companies winning on AI economics are not spending less - they are routing smarter. This is the practical playbook for model routing, SLM substitution, and building a FinOps-for-AI function that turns inference economics into a competitive advantage.

13 min read

Cybersecurity·Apr 3, 2026

What the Claude Code Source Leak Teaches Every Team Shipping AI Tools About Build Pipeline Security

Anthropic accidentally published Claude Code's entire source code via an overlooked source map in an npm package - the second time in 13 months. This article breaks down exactly what went wrong, why AI tooling is uniquely vulnerable to build pipeline oversights, and the five-point release engineering checklist that prevents it.

12 min read

AI Strategy·Apr 2, 2026

AI Is Entering Its Multi-Architecture Era. Here Is How Leaders Should Prepare.

A Turing Award winner just raised $1 billion to build AI that understands physical reality, not just text. The emergence of world models alongside LLMs signals a multi-architecture future. Leaders who understand both paradigms - and know where each excels - will have a decisive strategic advantage.

14 min read

AI & Infrastructure·Apr 1, 2026

Google's TurboQuant Makes AI 6x Cheaper to Run. Here's How Smart Companies Will Use That Advantage.

Google Research published a paper showing how to shrink AI memory usage by 6x with no loss in quality. While memory chip stocks dropped on the news, the real implication is far more interesting: TurboQuant is about to expand who can build, deploy, and compete with AI. The historical pattern - from coal to cloud computing - tells us exactly what happens next.

12 min read

Security·Mar 31, 2026

The Six-Generation Pattern Behind Every Email Attack - and How to Stay Ahead of Generation Seven

From executable attachments to business email compromise to AI-powered phishing, email security has followed a remarkably consistent pattern across six generations. Organizations that recognized each shift early built defenses before the damage hit. Generation Six - prompt injection against AI email agents - is underway, and the playbook for getting ahead of it already exists.

13 min read

AI Strategy·Mar 30, 2026

The Economics of On-Premises AI: A Decision Framework for Leaders Spending $10K+ Monthly on Cloud Inference

When DeepSeek proved that frontier-class AI could run on a fraction of the expected hardware, it changed the infrastructure calculus for every business running AI in production. Fourteen months later, 93% of enterprises are reevaluating where their AI workloads run. This is the decision framework - grounded in real TCO data - that separates strategic infrastructure choices from expensive defaults.

11 min read

Security·Mar 29, 2026

Deploying MCP-Connected Agents Securely: Lessons From the First Year in Production

MCP has become the standard protocol for connecting AI agents to business tools - databases, APIs, file systems, and more. As adoption accelerated through 2025, a handful of real-world incidents revealed the security patterns that matter most. Here is the practical playbook for deploying MCP-connected agents with confidence, informed by what actually went wrong and how leading teams prevent it.

10 min read

AI Strategy·Mar 28, 2026

The 5 Controls That Separate Reliable AI Agents From Costly Mistakes

Air Canada, Chevrolet, Meta, Nippon Life, and DPD all made headlines for AI agent incidents. In every case, the underlying model worked exactly as designed - the gap was in the accountability infrastructure around it. This article breaks down five documented cases, extracts the pattern behind each failure, and maps the five engineering controls that experienced teams implement before deployment. The regulatory landscape has caught up: California AB 316, the EU AI Act, and FTC Operation AI Comply all now enforce what smart implementers already practice.

14 min read

AI Strategy·Mar 27, 2026

How the Top 12% of Companies Are Getting Real ROI From AI Agents

PwC surveyed 4,454 CEOs and found a sharp divide: 12% reported AI that both grew revenue and cut costs, while most saw little return. Meta reports 30% higher output per engineer. Teams using AI agents save 10-12 hours per week. The difference is not which model you picked. It is whether anyone built the system around your actual business.

9 min read

AI & Marketing·Mar 26, 2026

ChatGPT Visitors Convert at 5x the Rate of Google Traffic. Here Is How to Capture That Advantage.

Visitors arriving from ChatGPT convert at 14% - roughly five times the rate of traditional Google organic traffic. Brands cited in AI Overviews earn 35% more clicks and 91% more paid engagement. The businesses capturing this advantage share a clear pattern: they have made their expertise visible and extractable across the platforms AI systems actually pull from. Here is the data-backed playbook.

12 min read