Skip to content

Blog

Thoughts on AI architecture, agentic systems, and engineering leadership.

Retrieval-Augmented Self-Improvement for Agentic Flows

Retrieval-Augmented Self-Improvement for Agentic Flows

Your AI Agent is Drifting. Here's a Pattern to Fix It Without Retraining.

April 16, 2026 10 min read
Agentic AIRAGMulti-AgentMLOpsProduction
Read more

The Agentic Angle on Anaplan: AI Just Got Access to Your Planning Model

An unofficial MCP server just gave AI agents real access to Anaplan. Seventy tools, OAuth2 auth, live data. Here's what agentic enterprise planning looks like today, what it cannot do yet, and why the gap is the most interesting part.

April 9, 2026 10 min read
Supply ChainAnaplanAgentic AIPlanningMCP
Read more

Simple Models, Complicated Reality: When to Stop Adding Variables

More variables, longer training runs, deeper feature engineering. The instinct is to keep adding complexity until the model matches the reality it's predicting. This is, reliably, the wrong instinct.

April 8, 2026 5 min read
Supply ChainModellingMachine LearningPlanning
Read more

The Last Mile of Supply Chain Visibility Isn't Logistics

You can track a pallet to within ten meters anywhere on the planet. What you can't reliably answer is whether the right data about that pallet is in the right system at the right time for the people who need to make decisions.

April 5, 2026 4 min read
Supply ChainData IntegrationVisibilityOperations
Read more

AI Agents for Supply Chain Exception Management

Every supply chain produces exceptions. Most organizations manage them through daily standups and escalation emails. There's a better way, and it involves agents that actually take action.

April 2, 2026 5 min read
Supply ChainAI AgentsTool UseOperations
Read more

We've Been Over This: Context Compaction, Agent Memory, and the Art of Re-Explaining Your Own Repo

You told the agent. It knew. It pushed. And now, eight commits later, it's asking which version control platform you use. A breakdown of why this happens and what nobody has solved yet.

March 30, 2026 8 min read
AI AgentsClaude CodeDeveloper ExperienceContextMemory
Read more

I Built a Multi-Agent AI Assistant for My Portfolio. Yes, It's Overkill. That's the Point.

Most portfolio chatbots are a system prompt wearing a blazer. I built one with 3 agents, 5 tools, and a real-time trace panel - because if you're going to claim you build agentic AI, maybe your website should prove it.

March 22, 2026 4 min read
Agentic AIMulti-AgentTool UsePortfolio
Read more

Why AI Agents Need Tool Use - Lessons from 3 Years in Production

Chatbots generate text. Agents take action. After running production AI agents with tool-use orchestration for over three years, here's what I've learned about what separates a real agent from a very expensive autocomplete.

March 20, 2026 4 min read
Agentic AITool UseLangChainProduction
Read more

S&OP Is a Data Integration Problem, Not a Meeting Problem

Most S&OP improvement programs focus on the process: better cadences, clearer RACIs, executive sponsorship. Very few fix what's actually broken: the data arriving late, incomplete, or quietly wrong.

March 14, 2026 4 min read
Supply ChainS&OPData IntegrationPlanning
Read more

Your Planning Model Is Lying to You (And It's Not the Model's Fault)

Connected planning tools promise a single version of the truth. What you get is a very polished version of the same arguments you were having before, now with better charts.

March 5, 2026 4 min read
Supply ChainPlanningS&OPData
Read more

Demand Forecasting Is an AI Problem Disguised as a Planning Problem

Everyone builds dashboards. Nobody fixes the signal. Here's why demand forecasting fails at the data layer long before it fails at the model layer.

February 20, 2026 4 min read
Supply ChainDemand ForecastingMachine LearningData
Read more
When Your AI Coding Agent Starts Speaking Russian

When Your AI Coding Agent Starts Speaking Russian

I asked GitHub Copilot to start a React app. It replied in Russian. Turns out, LLMs occasionally bleed between languages - and it's more fascinating than alarming.

February 14, 2026 3 min read
LLMsAI AgentsLanguage BleedDebugging
Read more

Copilot Studio Credits: The Price of Letting an AI Make a Decision

Microsoft Copilot Studio charges per action, per answer, per thought. I did the math so you can decide whether it's cheaper to hire a human to check your SAP inventory.

January 22, 2026 6 min read
Copilot StudioMicrosoftAI CostsEnterprise AI
Read more