Reliability

Layer 2. Silent failure is the AI failure mode. Evaluation replaces assertion, errors compound across chains, and monitoring without response capability is theater.

Reliability

The Normal Systems Recovered. The Agents Did Not.

Aug 11, 2026

•

9 min read

The Normal Systems Recovered. The Agents Did Not.

Last night 1.2 million queued files filled a disk and took down the data layer under an agent I'm building. We cleared it. Everything else came back. The agentic system came back different, and nothing on any dashboard said so.

Reliability

I Questioned a Distillation Claim. The Post Disappeared.

Aug 8, 2026

•

3 min read

I Questioned a Distillation Claim. The Post Disappeared.

Someone announced they had distilled a frontier model into something small enough to run locally, doing the same thing. One word in that claim is the word Theranos sold the world on.

Reliability

Your Sub-Agents Are Junior Developers. Who Writes Their Tickets?

Aug 8, 2026

•

4 min read

Your Sub-Agents Are Junior Developers. Who Writes Their Tickets?

Everyone is hiring AI engineers. The constraint on agentic delivery is test coverage, test coverage is a domain knowledge problem, and that points at a job nobody is posting for.

Reliability

They Could Swap AI Models in a Day. That Was the Problem.

Aug 8, 2026

•

5 min read

They Could Swap AI Models in a Day. That Was the Problem.

I ran a portability review this week on a team with better architecture than most. They had already swapped models twice with no incident. That turned out to be the worrying part.

Reliability

Fifty Sub-Agents at 95% Gives You a 7.7% Chance

Aug 8, 2026

•

5 min read

Fifty Sub-Agents at 95% Gives You a 7.7% Chance

Everyone spawning sub-agents believes more of them means finishing sooner. The arithmetic says the opposite, and the fix removes a component most people think they need.

Reliability

I Mapped My AI Software Lifecycle. Code Wasn't the Bottleneck.

Aug 7, 2026

•

6 min read

I Mapped My AI Software Lifecycle. Code Wasn't the Bottleneck.

AI made code generation fast. It did nothing for the two places the time actually goes, and one of them was quietly lying to me.

Reliability

The Normal Systems Recovered. The Agents Did Not.

I Questioned a Distillation Claim. The Post Disappeared.

Your Sub-Agents Are Junior Developers. Who Writes Their Tickets?

They Could Swap AI Models in a Day. That Was the Problem.

Fifty Sub-Agents at 95% Gives You a 7.7% Chance

I Mapped My AI Software Lifecycle. Code Wasn't the Bottleneck.

Chris Knows Production AI