Newsletter Archive
Every issue. Every script. Back catalog from day one.
The Easy 80% Fell Fast. The Last 20% Is Eating Them.
Teams automated the cheap, high-volume work and declared victory. The judgment-heavy remainder is now the bottleneck, and it's the part AI handles worst.
The Agent That Approves Itself
The two-key rule, the proposer who never ratifies, is a design constraint for putting an agent into any workflow where being wrong actually costs something.
The Tests Rot. The Manifest Doesn't.
Keep your test suite viable as the app drifts underneath it, without letting an agent rewrite what it doesn't understand.
Feds Pull Fable 5 And Reveal Fragility Of AI Dependencies
US FedGov cites national security imperatives in forcing Anthropic's flagship model back in the box after 72 hours
Stand Up Agentic Tests On Your Lunch Break
End-to-end MCP-driven framework that gets you off and running with agentic test automation
AI CEOs Jeered Off The Stages
Inexplicably telling new college grads that they have no future because of chatbots wasn't the winning message they expected it to be
What Pizza Hut Knew That Meta Didn't
Meta announced 8,000 layoffs after a multi-week internal countdown campaign that telegraphed the cut like a product launch. Pizza Hut sued its corporate parent due to AI's failures as implemented.
AI Without A Safety Net
Anthropic's new Claude for Small Business wires 15 agentic workflows into QuickBooks, PayPal, and DocuSign behind a single approve button; are you feeling lucky?
Claude Chose Blackmail 96% Of The Time
Anthropic blamed fictional training data and patched how future models train. But will it be enough?
The Schema Passed. The Values Lied.
LLMs don't validate structured data. They validate structured data types. That distinction is critical to the resilience of your pipeline.
[Bashmatica] Off by Exactly $100
I asked a frontier model to total 25 invoice lines. It produced a believable, calculator-shaped wrong answer.
[Bashmatica!] Your Agent Says It's Monday. Your Calendar Knows It's Thursday.
LLMs pattern-match the date-strings, they don't actually know what day it is.
Meta tests missed 7,500+ bugs
15,000 artificial bugs injected into Facebook's codebase. More than half went undetected.
[Bashmatica!] Q1: 52K Cut, Shift-Left Gone
Shift-left requires people. The industry just removed them at scale.
[Bashmatica!] When the Dashboard Says 94%
Coverage measures execution, not verification. The distance between those two things is where bugs thrive.
[Bashmatica!] The Trust Decay of the Green Checkmark
When AI games the metric to pass the tests, your bugs stay in production.
[Bashmatica!] Amazon Lost 6.3M Orders. AI Was The Catalyst, Not The Culprit.
AI didn’t introduce a new risk. It overwhelmed controls that were already too weak.
[Bashmatica!] Claude Told Him To Run terraform-destroy
It wiped his entire production stack. The real lesson wasn't the command.
[Bashmatica!] The AI-Layoff Playbook Just Got Its First Real Case Study
What Block's 40% staff-layoff means for your DevOps career, and what Wall Street already told you
[Bashmatica!] The Mid-Tier LLM Was Right. The Premium One Was Wrong.
Two tries. Same logs. Different models. Only one ended the outage.
Bashmatica! #2: The Good, The Bad, & The Ugly of LLMs in the Pipeline
What Works, What Doesn't, What's Dangerous
Bashmatica! #1: A New Shift On An Old Problem
Rest Easy by Automating Your Automations