The Hard Ones Don't Announce Themselves
Two articles in this series ended by pointing at a line between local and frontier. Neither built it — and the thing that draws it turns out not to be a model choice at all.
Why Are We Building the Car Before the Train?
America is racing toward general-purpose AI agents with open-ended agency. China is putting its industrial weight behind AI inside defined systems and workflows. Two centuries of transportation engineering suggest we are handing over the steering wheel before we have laid the track.
The Selection Is the Sentence
Every interaction should do more than one job without asking the user to do more than one thing. The Egyptian canon worked because the second reading was a rule, not a guess.
Tokens Are the New Clicks
The agentic inference-efficiency paradox, and what it does to the capex thesis.
The Travel Time Collapsed
A 1976 foraging model explains why the rational threshold for abandoning a software project just moved — and why that edge expires.
The Room Scales
What the threatened coworker was always going to teach us about agents — and the one auditing test that does not degrade as they get smarter.
The Wrong Knob
Qwen3.8-27B landed for local use this week. The first question everyone asks about it is the least interesting one you can ask — and on a graphics card, your VRAM already answered it.
Where the Equation Flips
The local-versus-cloud line moved while this was being written — and total parameters stopped being the axis it moves along.
When Intelligence Stops Going Vertical, And Spreads Horizontally
Why the next frontier of AI is not a bigger model, but a shared grammar for agents to reason, disagree, and learn together.
The Automatic Transmission Moment
Everyone says the programmer is moving up the stack. Almost no one asks where the stack stops... or what the people resisting it are actually protecting.
Cognition Is Not Free
The trilogy ended at trust. This is what comes next... the layer that decides what thinking is worth, and who pays for it.
The Honor System Does Not Scale
Agent coordination is not a trust problem you can solve with better prompts. It is a trust infrastructure problem — here is how we do it.
Legibility Is Not Enough
Being findable by machines was the easy part.
The AI Training Clause Every Commercial License Needs
What it has to do, what it can't do, and why every commercial license drafted in 2026 will need it.
The User Is Not the Customer Anymore
The interface layer is collapsing. The next buyer is a machine acting on someone's behalf, and most products are not legible to it.
I Built a Desktop App to Manage What Claude Code Leaves Behind
Claude Toolkit: artifact discovery, security scanning, permissions management, and settings repair for Claude Code power users.
Claude Code Has a Hygiene Problem Nobody's Talking About
The configuration sprawl nobody's managing is the technical debt of AI-assisted development.
Hybrid Search at 60 Million Documents
When Neither Keywords Nor Vectors Are Enough
Embedding 60 Million Documents
What Nobody Tells You About Vector Search at Scale
Why MCP Needs a Manifest: Solving the Configuration Chaos
From manual JSON copy-pasting to one-click installation, how mcp-manifest.json brings autodiscovery to the MCP ecosystem.
What Export Controls Actually Accelerated
The policy designed to constrain China's AI capability may be accelerating the form of efficiency that matters most.
Dig Deeper — Interactive
Read moreSet the market conditions, then design a road and whether trust federates.
Symbey Try it From: The Road Becomes the GateSet launch price, mass per kilowatt, economic life, build cost, and loss rate, then see break-even launch
Strikeline Try it From: The Number SpaceX Hasn't FiledSet the adaptation speed and watch whether the gap ever closes.
Mechanisms Try it From: The Locked DoorSet where sustained begins, how sensitive the hostility signal is, & where the line hits.
Symbey Try it From: The Purpose Test