Thinking on software identity, agentic development, and the engineering decisions that persist.
Karpathy's LLM Wiki and Nate B. Jones's Open Brain are making a fork visible in personal memory: compile at write time, or query at demand. The same fork runs through software identity. Here is how we are working on it one layer up.
Zendesk's AI Agents are moving to an MCP-native architecture. When the MCP Client ships, Ribo is the connector that anchors every automated reply to your system's declared intent, not to stale knowledge base articles.
Agents guess because you gave them nowhere to ask. A source of truth they can defer to, a channel they can use when the truth is incomplete, and a fleet of narrow downstream passes does the rest.
Every turn an agent spends rediscovering what a function is, who owns it, and what it promises is a tax on your latency, your token bill, and your error rate. The fix is not more context. It is identity upstream of discovery.
In 2025, the hard problem was finding enough context for agents. In 2026, the hard problem is finding less of it. When identity exists upstream, the new skill is slicing -- carving the smallest coherent cross-section of DNA that contains all the concerns for a given task.
The biggest fear CTOs have about identity engineering is that it requires pausing feature work. It doesn't. The strangler fig pattern lets you start with one service, prove the value in a single quarter, and expand without stopping the roadmap.
Identity engineering has real failure modes. Teams that adopt it badly end up worse than teams that never adopted it at all. This article names the three most common ways it goes wrong and the antidote for each.
Identity gaps have a dollar cost. Not a vague 'technical debt' cost -- an actual, calculable cost per quarter that shows up in onboarding delays, incident duration, review bottlenecks, agent rejection rates, and compliance overhead. Here is the model to calculate it for your own org.
Teams that declared what their software is supposed to be, then measured against it, keep reporting the same pattern: onboarding cuts in half, production outages drop to near zero, deploy frequency doubles. These are the closest real-world numbers we have for identity engineering, and they tell a consistent story.
Identity engineering has to fit into the workflow you already have or it will be rejected. This article maps exactly where identity declarations get created, reviewed, and updated within the sprint cycle, PR workflow, CI/CD pipeline, and incident response.
We use cookies to understand how you use ribo.dev and improve your experience.
Learn more in our Cookie Policy