The Blog
Writing on agentic AI.
Practical, honest pieces on building and designing agentic AI systems. Field notes from production work, written for engineers, architects, and technology leaders.
-
01
Enterprise · 24 min read
What Counts as Agentic AI, and What Does Not
Chatbots, workflows, and agentic AI are not the same thing. A working definition, the AGENT framework, the autonomy ladder, production gotchas, and a 10-question checklist you can run on Monday.
-
02
Enterprise · 9 min read
What to Automate First: A Leader's Framework for Agentic AI in 2026
The question that stalls most agentic AI programs is not "can we build an agent." It is "which process first." A leader's framework for choosing the first project: four scoring criteria, one hard gate (blast radius), a 2x2 you can draw on a whiteboard, and a worked example scoring three real processes.
-
03
Beginner · 12 min read
When to Use an Agent Framework vs Build It Yourself (LangGraph, CrewAI, and More)
The honest answer is not "always" or "never." Build the raw loop yourself until you hit a specific named pain, then adopt a framework to solve that pain, not on principle. A decision diagram, a map of the frameworks worth knowing (LangGraph, CrewAI, AG2, LlamaIndex, OpenAI Agents SDK, Pydantic AI, plus the AWS, Microsoft, and Google platforms) and what each is good at, and the signals that tell you which side of the line you are on.
-
04
Enterprise · 14 min read
Why Voice AI Agents Are Harder Than Chatbots
A working chatbot rarely survives the jump to a phone line. Why voice agents are harder: latency as a hard budget, barge-in, ASR errors, silence that means something, emotion, and real-time handoff.
- Aug 14 2026
The Hard Part of an AI Eval Set Is the Answer Key
An automated score said my retriever was failing badly. I read the failures by hand and found my own answer key was wrong. Here is what that cost, and what it did not fix.
- Jul 27 2026
What an AI Agent Actually Costs Per Resolved Customer Issue
Most AI contact-center business cases count too little cost and divide it by the wrong outcome. Price your agent per resolved customer issue, not per contained contact.
- Jul 23 2026
When the Small Language Model Isn't the Bottleneck
On an offline 350M assistant, model choice mattered, and then stopped being the limit. Several failures I blamed on the model started in the system around it.
- Jul 19 2026
Turn Up AI Autonomy Gradually, Like a Dial, Not an On/Off Switch
Autonomy is not an on/off property of an agent. It is a per-task setting you turn up gradually as the agent earns trust. Here is the dial, and how to move it.
- Jul 14 2026
The AI Agent Production Readiness Checklist
Twelve checks that decide whether an agent ships. Any red holds the launch until it is green or covered by a control, except four hard blocks that cannot be covered.
- Jul 05 2026
Forward Deployed Engineering: Why Agentic AI Doesn't Ship Itself
Forward Deployed Engineering feels new because AI vendors finally named it. The mid-2026 wave of billion-dollar commitments is the market admitting agentic AI does not ship itself.
- Jul 05 2026
LangGraph for Voice AI Agents: Strengths, Limits, and Production Risks
LangGraph is the stateful control-flow brain for a voice agent, not the real-time media layer. Where it fits, where it breaks, and how to compose it for production voice.
- Jun 23 2026
AI Agent Observability Is Not Just Logs
Logs tell you what the system did. Agent observability tells you why it decided wrong. Here is what to observe, what to evaluate, and what to instrument first.
- Jun 17 2026
Why Most Agentic AI Demos Fail in Production
A demo proves an agent can work once. Production asks whether it works repeatedly, safely, within budget, on messy inputs and broken tools, for real users. The four gates that decide whether a pilot ships, and why the demo-to-production gap is structural, not a bug to patch.
- Jun 17 2026
From Deflection to Resolution: How IVA Is Rewiring the Contact Center
Legacy IVR was built to deflect callers; intelligent virtual agents should be built to resolve them. Why that one shift re-prices who wins the contact center, what the 2026 funding wave is really buying, and why the moat is integration and CX design, not the model.