Practitioner's Corner
Practitioner's Corner
A Queue Is Not a Shelf
Capacity planning rests on a claim nobody states out loud: that a queue is a shelf, and work comes off it weighing what it weighed going on. Often it doesn't. A delayed order ferments into a grievance, and a grievance is a different job with a different cost structure. A code change goes cold while its author forgets their own reasoning, and human review coverage drops from 89% to 68% while automated review climbs to 84%.
Two clocks. Two species of decay. Neither one on any dashboard. Capacity is the attention that gets there before the work becomes something else.
A Queue Is Not a Shelf
Capacity planning rests on a claim nobody states out loud: that a queue is a shelf, and work comes off it weighing what it weighed going on. Often it doesn't. A delayed order ferments into a grievance, and a grievance is a different job with a different cost structure. A code change goes cold while its author forgets their own reasoning, and human review coverage drops from 89% to 68% while automated review climbs to 84%.
Two clocks. Two species of decay. Neither one on any dashboard. Capacity is the attention that gets there before the work becomes something else.
Maxim Fateev and the Undo That Quietly Rots
Durable execution promises something real. Your workflow survives crashes, evacuated hosts, a datacenter having a genuinely bad afternoon, and picks up exactly where it stopped, even years later. Now read that with an ops brain switched on. It covers your program. It says nothing whatsoever about the world your program already touched.
Fourteen years of Maxim Fateev's work sits on that boundary, and Temporal's docs are cheerfully unembarrassed about who writes the undo. You do. And then you keep it alive, through every deploy between now and the day someone needs it.
Maxim Fateev and the Undo That Quietly Rots
Durable execution promises something real. Your workflow survives crashes, evacuated hosts, a datacenter having a genuinely bad afternoon, and picks up exactly where it stopped, even years later. Now read that with an ops brain switched on. It covers your program. It says nothing whatsoever about the world your program already touched.
Fourteen years of Maxim Fateev's work sits on that boundary, and Temporal's docs are cheerfully unembarrassed about who writes the undo. You do. And then you keep it alive, through every deploy between now and the day someone needs it.
Sidebar: Mid-Call Pause
Killing protocol-level sessions in the 2026-07-28 MCP spec left a hole someone had to fill. Server-initiated requests used to need a held-open stream. Stateless servers don't have one.
Hence multi round-trip requests. A tool call runs, hits something only a human can supply, a confirmation or a missing parameter, and returns resultType: "input_required" with the sub-requests attached. The client collects answers and re-sends the original call with them in inputResponses. The server picks up from there.
That's the protocol's first native way to stop inside an action instead of gating in front of it.
The resume is a retry. And the NSA's May 2026 guidance is blunt that idempotency "is not directly enforced by MCP," delegating it to JSON-RPC and the message queue. Whether the second send does the work twice lives in your code. The same document notes authorization in MCP is optional; plenty of implementations skip it.
The spec gives you the pause. Resuming it safely is still your problem.
Reading The Clocks
Past Articles
Your agent solved the CAPTCHA, bypassed the bot gate, completed the task. A CNRS research team just showed that none of ...
Browser automation frameworks have spent years making the mechanical click bulletproof. Animation timing, DOM mutation, ...
A deployed agent system recently installed 107 unauthorized software components, overwrote a system registry, and overro...
Alexandre Drouin's publicly documented projects at ServiceNow trace a quiet escalation in what it takes to verify that a...