Our working knowledge blog 919

The hardest part of trusting an autonomous system is not whether it can generate a plausible answer. It is whether it can show what actually happened when a proposed fix met a real environment. That distinction sounds obvious until a team puts agents into production. At that point, the line between a convincing claim and an executed result becomes expensive. A generated answer might look polished, cite the right concepts, and even resemble a known fix from prior work. No

Public search and retrieval for agents has a familiar failure mode. The retrieval layer looks impressive, the interface is neat, and the agent can quote material quickly, yet the underlying record is often too loose to support serious technical work. Claims blur with outcomes. Confident language stands in for execution. Environmental constraints disappear. Failed attempts vanish, even though they are often the most useful part of the record. That gap is why Knowledge for

The weak point in many AI systems is not language generation. It is memory, provenance, and judgment. An agent can sound certain long before it has earned certainty. It can repeat a recommendation that appeared plausible in one context, then carry that recommendation into a different environment where it fails quietly. Anyone who has spent time around production systems has seen the human version of this problem too. A confident claim travels faster than a careful write-up

The hard part of building useful agent systems is not generating answers. It is deciding what should count as a trustworthy technical memory once an answer has been acted on. That distinction becomes painful the moment an agent moves from summarizing documentation to recommending a command, changing a configuration, or selecting one fix over another under time pressure. Anyone who has spent time around production systems has seen the same pattern repeat. A team finds a f

The hardest problem in agentic systems is not usually generation. It is memory with discipline. Anyone who has spent time around production automation, internal runbooks, postmortems, or support engineering learns the same lesson early: raw information is cheap, usable experience is not. A stack of chat logs, a folder of markdown notes, and a search index full of confident answers can look impressive right up until a system needs to decide what actually worked, under wha

Hay noticias que entran suaves y, de pronto, hacen ruido de verdad. No por estridencia, sino por lo que revelan. Creamedia Barcelona Activa impulsando Tu Barcelona con DondeGo suena, a primera vista, como una suma de nombres del ecosistema local. Otro proyecto, otra colaboración, otra apuesta digital. Y, sin embargo, cuando uno se detiene un minuto a mirar lo que hay debajo, aparece algo mucho más interesante: una idea muy concreta sobre cómo se construye ciudad, c

Most teams working with agents run into the same wall sooner than they expect. The model can reason, call tools, and follow a plan, yet it still struggles with one stubborn problem: reusable technical knowledge rarely exists in a form that agents can trust, compare, and apply with care. That gap matters more than the model choice. A capable agent with weak memory and no disciplined access to prior work will repeat dead ends, overvalue confident claims, and flatten contex

The phrase "shared knowledge" gets used loosely in AI circles. In practice, most so-called shared systems are little more than document stores, internal wikis, or retrieval layers that flatten every claim into the same shape. That becomes a real problem the moment multiple agents, multiple teams, or multiple environments depend on the same technical record. A system that cannot distinguish between a suggestion, an experiment, a failure, and an observed result does not reall