News & guides
A/B Testing Two Models Without Fooling Yourself
Comparing two models on live traffic sounds simple and usually is not. Sample sizes, paired designs, and the metrics that actually settle the question.
Active vs Total Parameters: The Number Spec Sheets Hide
A 2.8T model and a 27B model can be two-to-one apart on the figure that governs thinking. How to read parameter counts across the 2026 field.
· 9 min read
Agent Audit Logging: What to Record and What to Redact
When an agent does something surprising, the log is the only account of what happened. What a usable agent audit record contains, and what it must not.
· 9 min read
Agent Caching Strategies: Prompt, Tool and Result Layers
Agents resend the whole transcript every turn and re-read the same files repeatedly. Three caching layers fix that, and each has different keying rules.
· 9 min read
Agent Checkpointing: Saving Work a Long Session Can Lose
Long agent runs die halfway. What to checkpoint, where the boundaries belong, and why external side effects break the snapshot model entirely.
· 8 min readAgent Concurrency Control: Pools, Locks and Fair Slots
Unbounded agent spawning turns a fast run into a retry storm. Worker pools, semaphores, resource locks and the fairness problem nobody plans for.
· 9 min read
Agent Cost Attribution: Tracing Spend Back to a Cause
A provider bill tells you the total and nothing else. How to attribute agent spend to a team, a task and a tool call so you can act on it.
· 9 min readAgent Failure Modes: A Taxonomy Worth Memorising
Agents fail in about eight recognisable ways, and each one needs a different fix. A field guide to spotting them from a trace and knowing what to change.
· 9 min read
Agent Fan-Out Limits: How Wide Is Too Wide
Fan-out looks free until the orchestrator stops reading results properly. The four ceilings that cap parallel agents, and how to find yours before production does.
· 9 min readAgent File System Access: Designing the Blast Radius
The filesystem is an agent's most consequential tool surface. Path containment, read and write asymmetry, and why files are an injection vector.
· 8 min read
Agent Handoff Patterns: Passing Work Without Losing It
Every handoff between agents is a compression step. Four patterns for transferring control, what each one drops, and how to build a handoff packet.
· 9 min readAgent Loop Anatomy: The Twenty Lines That Run Everything
Every coding agent is the same short loop. Understanding its structure tells you where they fail and which parts are worth engineering.
· 8 min read
Agent Network Access Control: Egress Is the Real Risk
An agent with a shell and open network egress can exfiltrate anything it can read. How to scope outbound access without breaking the toolchain.
· 9 min readReady to try it?
Every model behind one key. Start free and upgrade when it sticks.