Advanced
Start here
Keep an agent’s permission attached to the action it actually takes
Connect authority, observability, containment, and ownership.
Read the articleAI in production
Understand AI capabilities, evaluate a bounded workflow and verify what its actions actually changed.
Advanced
Write an execution contract
Make an agent skill remember the action that did not finish
Describe the tool inputs, permitted actions, and failure handling.
Read the guideAdvanced
Evaluate the result
Did the incident agent fix checkout, or just say it did?
Build repeatable cases and an independent recovery gate.
Read the guideIntermediate
Account for the cost
Where an AI task’s tokens, time and cost go
Include retries and context limits in the operational picture.
Read the articlePut it into practice
Related AI news
News ·
GPT-6.1 Terra: from rough notes to a draft you can use
See where Terra fits writing, visual interpretation and app work, what its API prices cover, and how to compare it on a task with a visible result.
News ·
What Anthropic actually tested in its AI self-replication study
Anthropic ran a model across TV boxes and tested replication in a separate emulated fleet. Keeping those experiments apart reveals what the research actually shows.
News ·
LFM2.5’s small active count is only part of the local-agent job
An engineer’s local assistant may calculate less for each reply while still needing room for the whole model. Follow the tradeoff from loading to a useful answer.
More on ai in production
Explainer · 6 min read
MCP explained: how AI applications reach your tools and data
Follow a question from an AI application to the software holding the answer. MCP gives tools, resources and prompts a shared interface without replacing the systems behind them.
Explainer · 8 min read
Model quantization: how fewer bits make AI fit
Follow one weight through quantization, then see why model memory, speed and answer quality depend on more than a four-bit label.
Explainer · 13 min read
Your AI runs locally. What can it still send out?
A local model can keep inference on your computer. The surrounding app may still use cloud services, connected tools, telemetry and synced storage. Here is how those routes differ, and which choices actually change them.
Explainer · 8 min read
PHP2Shell: the real risk behind the AI-hacking alarm
An attacker may make a vulnerable PHP web worker run their code and reach its credentials. Follow the runtime fixes, later notices and separate investigation needed before calling the service recovered.