The Implementation Lab Notebook
A running log of real lessons from building & running AI systems, written as they happened. Each entry records what happened, why, & what changed as a result. The failures are in here too, dated & kept, because they carry the most transferable lessons. 26 entries & counting.
The Persistent Audit, Prototyped at Population One
What does a persistent, always-on audit look like when there's no team to run it — just one person?
The Distribution-of-Enforcement Audit — Why Rule Count Isn't What Collapses Systems
Is what makes a system collapse the number of rules, or who's actually enforcing each one?
The Stub Conversion That Ate Five Skills
Did consolidating five skills into pointers actually verify the destination held real content?
The Implementation Gap Ate the Implementation Specialist — the Vault Migration
Does the system running this business hold up under the same load its clients' systems face?
The Routing-README Pattern Reads as Novel in the Wild — Even Among Practitioners
Does a structural habit that feels basic from inside the work actually read as novel from outside it?
Pre-Marination Loops — Start Research Months Before You Need the Output
Can an agent start being useful before the moment it's actually needed?
Primary Role + Dual Entry Points — The Training Wheels Pattern
Does every AI-operator interaction need one fixed classification, or can it hold both?
Maintenance Automation Pattern — When You Fix It Twice, Automate It
Why does the same manual fix keep recurring, and when do I automate it instead?
Designed but never enforced — versioning protocol existed on paper for 3+ sessions with an empty _versions folder
Is the rule we designed actually running, or just written down somewhere?
"Review against" ≠ "read the file" — voice drift in AI-drafted content
Why did a drafted comment sound nothing like how Amanda actually talks?
"Fixed" Labels Create Single-Point Verification Failures
Was "fixed" actually fully verified, or just the one thing someone happened to check?
Source Attribution Drifts When Working From Secondary Summaries
Did the content I'm about to publish actually say what the primary source said?
Versioned Filenames Prevent Silent Overwrite Failures in Multi-Agent Systems
How do I stop silent overwrite failures when multiple AI tools touch the same files?
Platform Dependency Makes Your Own Data a Support Ticket
Can I actually get my own data and logic back out of the platform I built on?
Every Capability Jump Creates an Implementation Wave
What does a major model release actually mean for the organizations that have to adopt it?
Three-Tool Orchestration — The Pipeline Nobody Teaches
How do the different AI tools actually hand off to each other in one real build?
Manual Setup for "XP Points" — Learning vs Automating
Should I let AI set up the infrastructure, or learn it hands-on?
localStorage Is a Filing Cabinet Bolted to One Desk
What breaks when a single-device tool has to work across devices?
The Authorized Domain Error — Infrastructure Has Gates You Don't See
Why does sign-in work on desktop but break on mobile once it's actually deployed?
Chat History Wipe Erases an Entire Session
What happens when the platform itself loses a whole session with no warning?
The Competing Roadmap — Two Sources of Truth Means Zero
Where does the real plan live when chat keeps generating plausible-looking versions of it?
AI Capacity Estimation Ignores the Tools
Is capacity advice calibrated to what the tools now handle, or to doing everything by hand?
Stacked Deploys Make All Three Features Fail — Can't Isolate Cause
Why did three shipped features fail together, and which one caused it?
Silent Truncation Destroys 250 Lines of Coaching History
Why did detailed coaching history disappear from the file meant to hold it?
Platform Outreach References Multiple Podcasts for a Single-Podcast Host
Did the outreach email actually get the recipient's facts right before it went out?
AI Fabricates a Workplace Story That Sounds Exactly Right
Can I trust AI to generate my own work history from memory?
One implementation insight every other Tuesday.
What actually works when you ship AI in real organizations. Nothing you could get from a press release.