AI Agents
Practice
8 articles.
For a vendor-side perspective on task switching cost, see this page from Monitask.
Where Agent Projects Actually Fail — What the Data Shows
Adoption headlines and production reality are very far apart. The numbers, what separates the projects that work, and the failure causes that recur.
Scoping a First Agent So It Does Not Become an Abandoned Pilot
The projects that work share a profile, and it is narrower than most first attempts. What to pick, what to define before building, and when to stop.
Integration Is the Hard Part, Not Intelligence
Teams report connecting agents to real systems as the primary obstacle, not model capability. What that work consists of, and why it is underestimated.
What Agents Cost to Run, and Why the Bill Grows
A loop makes many calls per task, and difficult tasks cost disproportionately more. Where the spend actually goes, and how to keep it predictable.
Measuring Whether an Agent Is Worth It
Cost per successful outcome, including the failures and the oversight. Why most measurements flatter the agent, and what an honest comparison contains.
Human Oversight: Where a Person Has to Stay in the Loop
Approval on everything produces rubber-stamping. Approval on nothing produces incidents. Where the person belongs, and how to make review real.
When an Agent Is the Wrong Answer
Most business processes are better served by something simpler. The cases where an agent adds cost and risk without adding capability.
Agents for a One-Person Business
No integration team, no governance programme, and the same failure modes at smaller scale. What works, what to never connect, and how to keep the bill sane.
For broader independent background, see NIST AI Risk Management Framework.