Four Containment Surfaces, One Diagram: The Agent Stack Just Got Drawn
Compute, data, knowledge, identity. Four agent-control surfaces shipped reference implementations in one week. The architecture is no longer optional.
7 min read
Read more
The Thinking Wire
Governed Implementation
Speed and safety are not opposites. They are partners when governance is built in.
84 articles
Frameworks
The Anatomy of AI Agents: How Machines Sense, Think, and Act
From virtual assistants to autonomous vehicles, AI agents are everywhere. Learn the three-stage Sense-Think-Act architecture.
7 min read
Read more
From Pilot Purgatory to Production: A Practical Guide to AI Agents
Why two-thirds of organizations are stuck in pilot purgatory, and how to join the 8.6% that reach production with AI agents.
12 min read
Read more
Essays
- Essay
Five Primitives for Compounding with AI
Eugene Yan's five primitives for compounding with AI: context, config, verification, delegation, loop closure. Two of them are the ones most teams skip.
7 min read Read more - Essay
AI Value Is Leaving the Model Layer. Here Are the Three Places It's Going.
Three independent analyses converged on the same thesis this week. Model performance is becoming table stakes. The new moats live elsewhere.
8 min read Read more - Essay
Governance Just Shipped as a Product Feature. Three Times. In One Week.
Anthropic, Figma, and Vercel each turned governance into a product surface in the same week. Procurement now has comparable line items.
8 min read Read more - Essay
Terminal-Bench 2.0: The Harness Is the Procurement Decision Your CIO Isn't Making
Same model, different harness, 4.5-point swing. Terminal-Bench 2.0 confirms harness selection is a procurement variable, not an implementation detail.
6 min read Read more - Essay
Three Harness Architecture Decisions Buyers Are Making Without Knowing It
Last week the harness was a SKU. This week it is an architecture. Three decisions buyers are signing off on, often without realizing it.
8 min read Read more - Essay
Hub-Spoke Costs 4x More for Coding, and Loses. The Multi-Agent Question Got Measured.
Krishnan measured multi-agent architectures: coding favors solo, reasoning favors markets, hub-spoke costs 4x more. Procurement now has a number.
6 min read Read more - Essay
Three Teams, Three Skills Designs: The Architecture Debate Just Went Public
Vercel, Perplexity, and Addy Osmani published three incompatible Skill designs in six weeks. The disagreement reveals where governance must hold.
8 min read Read more - Essay
The Substrate Adapts: Design Systems, Docs, and Judgment in the Agent Era
Three writers, three domains, one shift in a single week: static substrates do not survive agents. Tokens, ADRs, and judgment have to move with them.
7 min read Read more - Essay
The Week Cloud Vendors Said: Stop Building Your Own Agent Harness
OpenAI on Bedrock, Mistral Vibe, AWS Neuron Agentic. In seven days, agent runtimes became a vendor purchase, not an in-house build.
7 min read Read more - Essay
Design Tools That Bypass the Constraint Layer
AI design tools learned code primitives, not your design system. The fix is not designer discipline — it is constrained generation.
7 min read Read more - Essay
Three Hyperscalers Shipped the Same Idea in One Week. Workspace Is the Control Plane.
OpenAI, Google, and Adobe each shipped a workspace-as-agent-environment play in seven days. The platform layer for agent governance is being defined right now.
6 min read Read more - Essay
The Ungoverned 60%: Why Context Engineering Is Becoming a Cross-Functional Discipline
Capability is no longer the bottleneck. The ungoverned 60% of product knowledge outside the codebase is. Three writers converge on the same fix.
8 min read Read more - Essay
Your Agent Permission Model Works 40% of the Time
ManyIH-Bench shows frontier models resolve instruction hierarchy at ~40% when sources exceed four tiers. Your permission model runs on a 40% base.
8 min read Read more - Essay
When Designers Code Again, the Org Chart Moves
LukeW says designers should code again. The real story is governance - when AI collapses the handoff wall, the boundary moves and ownership shifts.
6 min read Read more - Essay
Orchestration Is the New Governance Layer
Four April 2026 signals converge: role separation — planner, worker, validator — is replacing prompt engineering as the unit of AI governance.
9 min read Read more - Essay
Your Harness, Your Memory — And the Moat You Can't Take With You
If memory is the moat and memory lives in the harness, a vendor-owned harness is a vendor-owned moat. The exit question most teams can't answer.
7 min read Read more - Essay
Claude Managed Agents: When the Harness Becomes a Vendor Product
Anthropic turned the agent harness into a SKU. Build-vs-rent is now a governance decision, not a technical one. Here is the split.
7 min read Read more - Essay
The Agent Production Stack Now Has Governance Built In. But Whose Governance?
Anthropic decouples brain/hands/session. Claw-Eval grades safety alongside completion. Infrastructure catches up to capability.
9 min read Read more - Essay
SDD at Enterprise Scale: A Governance Problem in Tooling Clothes
Spec-driven development's 7 enterprise adoption barriers are organizational, not technical. Here is what the tooling vendors won't tell you.
11 min read Read more - Essay
Structured Prompting Is a Governance Mechanism for AI Code Reasoning
Semi-formal reasoning forces AI agents to show verifiable evidence before making code judgments. The real value is auditability, not accuracy.
9 min read Read more - Essay
The Intent Layer: Why the Best AI Governance Is the Kind Nobody Notices
Spacelift Intent, Google Skills, Claude Code harnesses — the pattern is clear. Effective governance is invisible, embedded in tooling, not bolted on.
8 min read Read more - Essay
The Spec Layer: Five Independent Teams Discovered the Same Agent Governance Architecture in One Week
Matt Rickard, Google, GitHub, Anthropic, and Kent Beck converge on specs as the primary governance mechanism for AI agents.
14 min read Read more - Essay
Your AI Will Hallucinate. Build the System That Catches It.
Hallucinations are intrinsic to probabilistic AI. The discipline that matters is not elimination but governance: layered detection, validation, and control.
9 min read Read more - Essay
Free Software in the Agent Era: A Governance Reckoning
AI agents let non-technical users modify and redistribute software. Open-source is now a governance decision, not just a technical one.
9 min read Read more - Essay
Design-First Is the Cheapest AI Governance You Can Buy
AI assistants collapse design into code. A five-level framework forces the decisions back into the open. But agentic AI needs more.
8 min read Read more - Essay
Three Roads to Governed Autonomy: What PostHog, Ossature, and Anthropic Converge On
PostHog built 44 tools before choosing MCP. Ossature added verification loops. Anthropic shipped a safety classifier. All constrain the environment.
9 min read Read more - Essay
Three Teams, One Pattern: What PM, Security, and Infrastructure All Built for Agent Governance
A PM platform, a security team, and an infra provider independently built governed AI agents. They converged on four identical patterns.
8 min read Read more - Essay
When Visa Extends Your Agent Payment Protocol: What MPP Means for Governed Commerce
Stripe and Paradigm launched MPP with Visa, Mastercard, and both AI labs. The protocol is live. The governance is not.
8 min read Read more - Essay
Three Views on Governing AI Code: Types, Specs, and Verifiable Trust
Vint Cerf says trust is infrastructure. Lean 4 says types are proof. A Haskell expert says specs become code. All three are right.
10 min read Read more - Essay
The Containment Convergence: Why NVIDIA, Docker, and OpenAI All Shipped Agent Sandboxes This Week
Three companies shipped agent containment in one week. The pattern is identical: YAML policies, egress proxies, credential isolation.
12 min read Read more - Essay
What 220 Controls Teach About Building AI Governance Frameworks
GitLab cut SOC controls by 58% with a custom framework. AI can now rewrite GPL code in days. Both stories reveal how governance actually works.
9 min read Read more - Essay
Agent Specs Are Governance Artifacts
Three independent frameworks converge on the same conclusion: agent specs are not documentation. They are auditable, enforceable governance infrastructure.
12 min read Read more - Essay
Skills Are Not Replacing Agents. They Are Making Agents Governable.
The Agent Skills standard solves what monolithic agents never could: modular, auditable, version-controlled AI capabilities.
7 min read Read more - Essay
The Governance Layer That Was Already There
Three practitioners independently rediscovered the same truth: AI agents need engineering discipline, not new frameworks.
7 min read Read more - Essay
The Architecture of Agent Trust: Why Boundaries Beat Instructions
Reliable AI agents come from environmental constraints, not better prompts. Three independent sources converge on the same architectural principle.
9 min read Read more - Essay
The Cage Pattern: What Agent Fleet Governance Actually Looks Like
Three companies running AI agents at scale converged on the same principle: maximum autonomy inside structural constraints.
10 min read Read more - Essay
Agents That Learn: The Missing Layer in AI Systems
Most AI agents forget everything between sessions. Learn how runtime learning transforms agents from tools into teammates.
12 min read Read more - Essay
Memory for AI Agents: Benchmarks, Architectures, and a Surprising Discovery
A simple filesystem outperforms sophisticated memory solutions. Discover what benchmarks reveal about memory architectures for AI agents.
15 min read Read more - Essay
AI Agents for Product Managers
How to use AI that acts and free up time for what matters. Use cases and tools for PMs without code.
10 min read Read more - Essay
Genie: How Uber Organized Data for AI That Works
Insights from Uber's Gen AI on-call copilot. RAG vs fine-tuning, Spark pipeline, and the quality secret.
12 min read Read more - Essay
How to Write Specs for AI Agents
The definitive guide for specifications that work. 5 principles tested by Google and GitHub engineers.
12 min read Read more - Essay
Agentic AI: When Machines Don't Just Chat, But Act
The difference between AI that responds and AI that acts. How agentic systems transform expectations and deliver 4x productivity.
9 min read Read more
Points of View
- Point of View
Cloud Next 2026 Field Notes: Spec-Driven Development, From the Room
Google engineers walked through spec-driven development on stage at Cloud Next 2026. The methodology is not new. The public commitment is.
7 min read Read more - Point of View
The Governed Runtime Goes Cloud-Native: Substrate Eats the Stack
Bedrock Managed Agents, Claude creative connectors, and AI-friendly APIs. Three signals, one direction. Governance is no longer a layer. It is the platform.
7 min read Read more - Point of View
Three Voices, One Verification Gap: The Harness Is Cross-Discipline Now
Three independent practitioners reached for the same idea this week using different vocabularies. AI velocity breaks unless verification keeps up.
7 min read Read more - Point of View
Symphony Names the Control Plane
OpenAI just open-sourced a spec calling Linear-class issue trackers the control plane for coding-agent fleets. The vocabulary fight has started.
6 min read Read more - Point of View
The Bounding Box Comes From Infrastructure, Not From a Prompt
Docker's microVM architecture makes the argument concrete. An LLM deciding its own security boundaries is not a security model. Infrastructure is.
6 min read Read more - Point of View
OpenAI's Agents SDK Learned to Run. It Still Cannot Be Governed.
OpenAI's SDK update adds sandboxes and memory. It solves execution. It does not solve governance. The attack surface grew; the controls did not.
8 min read Read more - Point of View
The Manifesto Arrives: Auditing McKinsey's 12 Themes
McKinsey names trust and guardrails for the first time. The sixth chapter in our tracking series. What the manifesto still leaves out.
8 min read Read more - Point of View
Thoughtworks Discovered Governance-as-Code. We Have Been Building It.
Fowler's team frames AI instruction files as developer tooling. The real move is organizational governance. Here is where their pattern stops and ours begins.
8 min read Read more - Point of View
Making Your Product Agent-Ready Is a Governance Decision
AI agents are selecting vendors autonomously. Optimizing for agent discovery without governing what you expose is a liability.
7 min read Read more - Point of View
OpenAI Built Governance Into the Platform. That Is Not Enough.
OpenAI built governance into their agent platform. That solves the containment problem. It does not solve the governance problem.
10 min read Read more - Point of View
McKinsey's Two Paths Are One Question They Won't Ask
McKinsey frames enterprise architecture as incremental vs comprehensive. The real choice is governance maturity. Fifth in our tracking series.
9 min read Read more - Point of View
Your Governance Reviews Quarterly. Your Models Update Every 5 Hours.
Cursor ships model checkpoints every 5 hours. OpenAI's own training lags its spec. Governance built for quarterly review cannot survive continuous deployment.
9 min read Read more - Point of View
Design Systems Just Became AI Governance Infrastructure
Figma's MCP beta lets agents write to the canvas. Design systems are no longer style guides. They are constraint layers for autonomous software.
6 min read Read more - Point of View
McKinsey's Agent Factory Prescription Is Missing Its Own Diagnosis
McKinsey prescribes agent factories to fix 80% AI failure. They left out governance. Again. The fourth chapter in a pattern we have been tracking.
8 min read Read more - Point of View
OpenAI's Governance Infrastructure Goes Public
OpenAI published its Model Spec and Safety Bug Bounty. What these artifacts reveal about treating governance as product infrastructure.
9 min read Read more - Point of View
McKinsey's Three Diagnoses: How a Consulting Firm Slowly Discovers Governance
McKinsey went from measuring AI wrong to calling it a design problem to using the word governable. The pattern reveals more than any single article.
8 min read Read more - Point of View
Design Without Governance Is Decoration
McKinsey says AI's scaling problem is a design problem. They're half right. Design is the interface layer of governance.
9 min read Read more - Point of View
Everyone Has 30% AI Code. Nobody Knows Who Governs It.
Uber and Stripe built governance into infrastructure. Microsoft hired a quality head. The pattern reveals what most companies are missing.
12 min read Read more - Point of View
Your Agent's Personality Is a Governance Layer
OpenAI published four personality profiles for AI agents. They missed the point. Personality is fleet-wide behavioral governance, not prompt cosmetics.
10 min read Read more - Point of View
McKinsey Measured the Wrong Thing
Executives self-report 16-45% AI gains. Controlled trials show 19% slowdown. The perception mismatch is not noise. It is missing infrastructure.
8 min read Read more - Point of View
Agent Frameworks Are Free Because You Are the Product
Hyperscalers give away agent SDKs to sell runtime. The real contest is governance: security, evaluation, context control. Bet on that layer.
7 min read Read more - Point of View
When the Spec Is the Product, Who Governs the Spec?
Spec-driven development compresses PM cycles. It also turns every unclear requirement into a production risk nobody reviews.
9 min read Read more - Point of View
The Governance Outsourcing Problem: What OpenAI's Consulting Alliances Actually Mean
When consulting firms deploy your AI agents, they also define your governance. Enterprises need to decide who owns the rules.
8 min read Read more - Point of View
Your Organization Is a Tactical Tornado
LLMs don't just write tactical code. They turn entire organizations into tactical tornadoes. The fix isn't better code review.
9 min read Read more - Point of View
Governance Is a UX Problem: What Apple's Agent Taxonomy Reveals
Apple mapped 55 UX features for AI agents. The finding most teams miss: governance that users cannot see is governance that does not work.
9 min read Read more - Point of View
The Containment Pattern: Four Approaches to Sandboxing AI Agents in Production
Cursor, Docker, Zenity, and Entire shipped four distinct containment layers in one week. The shift from approval fatigue to trust boundaries is here.
11 min read Read more - Point of View
The Rise of Agentic Platforms: Why Governance Is the Product
Bounded autonomy is the right design target for AI agents. Platform engineering is the governance layer most organizations already have.
10 min read Read more - Point of View
The Governance Gap in AI Output Quality: What Generic Design Reveals
Anthropic published a guide to fix generic AI designs. The real lesson: output quality requires the same governance as safety and compliance.
9 min read Read more - Point of View
The Thinking Gap Is a Governance Gap
Every concern about AI code generation maps to a governance failure, not a technology deficiency. The question was never whether to use AI.
11 min read Read more - Point of View
Your Style Guide Is a Governance Layer
Product teams encoding brand rules into CLAUDE.md are doing governance-as-code. Content ops proves the pattern.
9 min read Read more - Point of View
The Developer's Real Edge Isn't Business. It's Governance.
AI is reshaping software careers. The advice to 'learn business' misses the point. The premium is in governance-aware architecture.
9 min read Read more - Point of View
Product Discovery in the AI Era: When Building Is No Longer the Hard Part
AI compresses feasibility, viability, and usability risk. Desirability becomes the only differentiator. What changes for product teams.
9 min read Read more - Point of View
AI Agent Orchestration: From Prototype to Production
Why AI agents fail in production and how orchestration fixes it. Temporal, Conductor, and LangGraph compared.
12 min read Read more - Point of View
AI in Design Workflows: What Atlassian, Meta, Tesco, and Faire Are Doing Differently
How leading companies integrate AI into design processes with templates, instruction files, plugins, and internal chatbots.
10 min read Read more - Point of View
Against Cleverness: Design Principles for AI in Complex Systems
Why conservative design beats clever automation. Lessons from nuclear engineering for building AI systems that fail gracefully.
10 min read Read more - Point of View
Claude's Constitution: Lessons for Enterprise AI Governance
Analysis of Anthropic's 23,000-word framework and how to apply its principles to corporate AI governance.
12 min read Read more - Point of View
Google UCP: Universal Commerce Protocol for Agentic AI Shopping
Google UCP enables AI agents to complete purchases conversationally. See how retailers integrate and what it means for e-commerce.
8 min read Read more - Point of View
AI Agents: The New Frontier of the Job Market
With salaries reaching $4,000/month, AI agent specialists are among Brazil's most sought-after professionals. But what does this mean for your company?
6 min read Read more - Point of View
One Year of Agentic AI: Six Lessons from Those Doing the Work
After a year of agentic AI projects, clear patterns emerge. Six fundamental lessons for capturing real value with autonomous agents.
8 min read Read more
Explore Other Perspectives
Start Your Implementation
Building and deploying AI with governance baked in from day one.
Start Your Implementation