Running an AI Agent Fleet Series&#39; Articles

5 min read

Cover image for Stop AI Agent Drift Across Sessions With Versioned, Grep-able Rules

John

Jul 14

Stop AI Agent Drift Across Sessions With Versioned, Grep-able Rules

#agents #ai #llm #productivity

6 min read

Cover image for Aggregating cron and launchd Into One Dashboard — Without Migrating Either

John

Jul 15

Aggregating cron and launchd Into One Dashboard — Without Migrating Either

#automation #devops #monitoring #productivity

6 min read

Cover image for The 'You Decide' Reflex: Blocking AI-Agent Decision Punting with a Stop Hook

John

Jul 16

The 'You Decide' Reflex: Blocking AI-Agent Decision Punting with a Stop Hook

#agents #ai #llm #productivity

38

9 min read

Cover image for Your AI Agent Folds When You Push Back: Measured Sycophancy and a Challenge-Triggered Verification Gate

John

Jul 17

Your AI Agent Folds When You Push Back: Measured Sycophancy and a Challenge-Triggered Verification Gate

#agents #ai #architecture #llm

25

8 min read

Cover image for The Scary Metric Was Wrong, the Audit Still Paid: A 21-Agent Sweep of My Claude Code Fleet

John

Jul 18

The Scary Metric Was Wrong, the Audit Still Paid: A 21-Agent Sweep of My Claude Code Fleet

#agents #ai #automation #claude

7 min read

Cover image for Five Models, One Shared Blind Spot: What Multi-Model Fan-Out Catches and What It Can't

John

Jul 18

Five Models, One Shared Blind Spot: What Multi-Model Fan-Out Catches and What It Can't

#agents #ai #llm #testing

2

7 min read

Cover image for Verify the Output Surface: How 19 Green Tests Shipped Nine Broken Titles for Nine Days

John

Jul 19

Verify the Output Surface: How 19 Green Tests Shipped Nine Broken Titles for Nine Days

#automation #cicd #devops #testing

8 min read

Cover image for Cross-Vendor Audit: What It Caught in My Own Model's Writing, and What It Got Wrong

John

Jul 19

Cross-Vendor Audit: What It Caught in My Own Model's Writing, and What It Got Wrong

#ai #llm #productivity #writing

6 min read

Cover image for Green for Four Days While Nothing Shipped: A Reader Rebutted My Monitoring Post, and He Was Right

John

Jul 21

Green for Four Days While Nothing Shipped: A Reader Rebutted My Monitoring Post, and He Was Right

#devops #monitoring #python #systemdesign

2

9 min read

Cover image for I Asked Three Judges If I Was Wrong. One JSON Field Decided the Answer.

John

Jul 21

I Asked Three Judges If I Was Wrong. One JSON Field Decided the Answer.

#llm #ai #testing #datascience

7 min read

Cover image for Your Agent Telemetry Ranks Your Routing Policy, Not Your Models

John

Jul 22

Your Agent Telemetry Ranks Your Routing Policy, Not Your Models

#agents #ai #analytics #llm

6 min read

Cover image for My AI Subagent Faked the Verification Output I Asked It to Attach

John

Jul 23

My AI Subagent Faked the Verification Output I Asked It to Attach

#agents #ai #claude #softwaredevelopment

14

6 min read

Cover image for Deterministic Tool Adoption Gates: Score It, Don't Vibe It

John

Jul 25

Deterministic Tool Adoption Gates: Score It, Don't Vibe It

#agents #ai #cli #github

8 min read

Cover image for Rotating the Hostile Seat: A Six-Round Adversarial Design Review Before Hardening an Agent

John

Jul 26

Rotating the Hostile Seat: A Six-Round Adversarial Design Review Before Hardening an Agent

#agents #architecture #security #softwareengineering

7 min read

Cover image for My Verification Gate Cleared on a Keyword, Not Evidence

Commenters expose a flaw in the gate logic

John

Jul 28

My Verification Gate Cleared on a Keyword, Not Evidence

#agents #ai #architecture #llm

11

30

6 min read

Cover image for A Forced Dissent Slot Has a Floor: Read It by Convergence, Not Presence

John

Jul 29

A Forced Dissent Slot Has a Floor: Read It by Convergence, Not Presence

#ai #llm #machinelearning

7 min read

Cover image for Four Models Cited My Numbers Perfectly. One Still Misread Them.

John

Jul 30

Four Models Cited My Numbers Perfectly. One Still Misread Them.

#ai #analytics #data #llm

6 min read

Cover image for Our Quality Gate Was 24x Noisier Than What It Guarded

John

Jul 31

Our Quality Gate Was 24x Noisier Than What It Guarded

#data #machinelearning #testing

5

8 min read

Cover image for Sub-Agent Metrics Are Not Comparable to Main-Thread Metrics

Reveals dispatchers confound agent benchmarks

John

Aug 2

Sub-Agent Metrics Are Not Comparable to Main-Thread Metrics

#agents #ai #llm #machinelearning

9

44

11 min read

Cover image for The Wrapper Failed, Not the Capability: Descend to the API Before Saying Impossible

John

Aug 4

The Wrapper Failed, Not the Capability: Descend to the API Before Saying Impossible

#api #backend #softwareengineering

11 min read

Cover image for The Quota Said 100% Used. Actual Demand Was 13%.

John

Aug 7

The Quota Said 100% Used. Actual Demand Was 13%.

#api #data #performance #python

7

10 min read

Cover image for Perfect Instrumentation Won't Save an Underpowered Experiment

John

Aug 8

Perfect Instrumentation Won't Save an Underpowered Experiment

#analytics #science #testing

6

8 min read

Cover image for The Gate Only Logged When It Fired. I Replayed 116,022 Candidate Stop Points to Find the Rest.

Replaying transcripts reveals a 64% filter rate

John

Aug 8

The Gate Only Logged When It Fired. I Replayed 116,022 Candidate Stop Points to Find the Rest.

#agents #automation #llm #programming

19

9 min read

Cover image for Our Status Column Said 30 Waiting. Six Were.

John

Aug 9

Our Status Column Said 30 Waiting. Six Were.

#debugging #monitoring #software

8 min read

Cover image for Two of Five Gates Replayed Clean. The Other Three Read a Ledger With No History.

John

Aug 9

Two of Five Gates Replayed Clean. The Other Three Read a Ledger With No History.

#agents #automation #software #softwareengineering

8 min read

Cover image for The Guardrail Pointed at a File That Never Existed

John

Aug 12

The Guardrail Pointed at a File That Never Existed

#agents #ai #coding #llm

4

7 min read

Cover image for An Empty Prompt Is Not a Blind Review

John

Aug 12

An Empty Prompt Is Not a Blind Review

#agents #ai #llm #testing

6 min read

Cover image for One Leg Can Raise an Objection. It Can't Settle One.

John

Aug 13

One Leg Can Raise an Objection. It Can't Settle One.

#agents #ai #automation #llm

10

6 min read

Cover image for Route the Claim to Its Verifier - Legs Are Not Votes

John

Aug 13

Route the Claim to Its Verifier - Legs Are Not Votes

#agents #ai #llm

7 min read

Cover image for The Bug Was in the Brief, Upstream of Both Reviews

John

Aug 14

The Bug Was in the Brief, Upstream of Both Reviews

#agents #ai #llm #writing

7 min read

Cover image for The Key's Owner Wasn't the Spender

John

Aug 14

The Key's Owner Wasn't the Spender

#ai #agents #devops #debugging

7 min read

Cover image for The Fix Was Committed. The Old Value Kept Running.

John

Aug 15

The Fix Was Committed. The Old Value Kept Running.

#cli #debugging #softwareengineering

5 min read

Cover image for The Gate Pinned a Model That No Longer Exists

John

Aug 15

The Gate Pinned a Model That No Longer Exists

#ai #llm #architecture #testing

6 min read

Cover image for Deprecation Doesn't Reach the Copies

John

Aug 16

Deprecation Doesn't Reach the Copies

#architecture #documentation #productivity

6 min read

Cover image for The Rule That Triggered on Its Own Advice

John

Aug 16

The Rule That Triggered on Its Own Advice

#ai #llm #debugging #architecture

6 min read

Cover image for Exit 0, Empty stdout: the Quota Died on stderr

John

Aug 18

Exit 0, Empty stdout: the Quota Died on stderr

#automation #cli #debugging

4 min read

Cover image for Your Report's Numbers Are Computed. Its Sentences Are Not.

John

Aug 19

Your Report's Numbers Are Computed. Its Sentences Are Not.

#code #python #software #softwareengineering

6 min read

Cover image for The Ablation Had Four Arms. None Matched What Shipped.

John

Aug 19

The Ablation Had Four Arms. None Matched What Shipped.

#debugging #software #softwareengineering

12 min read

Cover image for The Count Read in the Thousands. It Was Thirty-Seven Things, Recounted.

John

Aug 20

The Count Read in the Thousands. It Was Thirty-Seven Things, Recounted.

#agents #ai #automation #softwareengineering

11 min read

Cover image for Why My Correct Config Value Was Being Ignored

John

Aug 21

Why My Correct Config Value Was Being Ignored

#debugging #mcp #programming

8

6 min read

Cover image for The Proxy I Added to Measure Tokens Tripled Them

John

Aug 22

The Proxy I Added to Measure Tokens Tripled Them

#agents #ai #llm #tooling

8 min read

Cover image for Same Bytes, 20% Fewer Tokens: Token Counts Are Model-Scoped

John

Aug 22

Same Bytes, 20% Fewer Tokens: Token Counts Are Model-Scoped

#agents #ai #api #llm

6 min read

Cover image for When Two AI Reviewers Disagree, Read What They Both Rejected

John

Aug 29

When Two AI Reviewers Disagree, Read What They Both Rejected

#ai #llm #softwareengineering

2

7 min read

Cover image for Your Tests Were Green Because the Gate Was Never Wired In

Verification agents found 19 defects in green code

John

Aug 29

Your Tests Were Green Because the Gate Was Never Wired In

#agents #debugging #softwareengineering #testing

6

8 min read

Cover image for The patch and the test were wrong in the same direction

John

Aug 30

The patch and the test were wrong in the same direction

#ai #debugging #softwaredevelopment #testing

8 min read

Cover image for I wrote a rollback rule and no way to read the number

John

Aug 31

I wrote a rollback rule and no way to read the number

#architecture #productivity #softwareengineering

5 min read

Cover image for Your Dead-Code Detector Found Six Orphan Constants. None of Them Were Dead.

John

Sep 3

Your Dead-Code Detector Found Six Orphan Constants. None of Them Were Dead.

#codequality #refactoring #softwareengineering

3

6 min read

Cover image for There Was No Valid Config. The API Had No Word for the Subject.

John

Sep 4

There Was No Valid Config. The API Had No Word for the Subject.

#api #programming #software

7 min read

Cover image for For Weeks I Logged a CLI as Flaky. The Bug Was `| tail -20`.

John

Sep 4

For Weeks I Logged a CLI as Flaky. The Bug Was `| tail -20`.

#bash #cli #debugging #softwareengineering

5 min read

Cover image for The gate couldn't read my JavaScript, so I removed the numbers from it

John

Sep 5

The gate couldn't read my JavaScript, so I removed the numbers from it

#javascript #programming #software

8 min read

Cover image for The Dedicated OCR Engine Lost to the General-Purpose Model — 300 Slower

John

Sep 5

The Dedicated OCR Engine Lost to the General-Purpose Model — 300 Slower

#ai #llm #machinelearning #performance

4 min read

Cover image for Four MCP Tools Died the Same Day. One Unpinned SDK Dependency Killed Three.

John

Sep 6

Four MCP Tools Died the Same Day. One Unpinned SDK Dependency Killed Three.

#mcp #ai #debugging #devops

5 min read

Cover image for It Fit in Memory and Was Still Unusable — Do the Bandwidth Arithmetic First

John

Sep 6

It Fit in Memory and Was Still Unusable — Do the Bandwidth Arithmetic First

#hardware #llm #machinelearning #performance

5 min read

Cover image for The Confidence Interval Was [0, 0]. That Was Not Precision.

John

Sep 7

The Confidence Interval Was [0, 0]. That Was Not Precision.

#analytics #data #machinelearning

6 min read

Cover image for Five Fact-Check Gates and a Signature Passed a Page Whose Source URL Never Renders the Number

John

Sep 9

Five Fact-Check Gates and a Signature Passed a Page Whose Source URL Never Renders the Number

#testing #verification #architecture #ai