Softlander blog perspectives published in July 2026.
Long-running agent work cannot rely on a transcript and hopeful retries. Explicit state makes progress, authority, recovery and completion inspectable across failures.
Accept an agent against outcome distributions, critical invariants and recovery evidence—not one scripted response or an average score.
Model usage is one line in a wider cost stack that includes integration, evaluation, control, human work, incidents, change and exit.
Agility is not a property of the algorithms you chose. It is a property of how your code, your data formats and your protocols name them—and whether anything can change without a coordinated rewrite.
Reviewable agent decisions depend on a small chain of artefacts connecting workflow purpose, authority, configuration, evidence, findings, controls and ownership.
Agents earn their complexity only when a workflow needs bounded adaptation. A disciplined no-agent test can reveal when rules, search or ordinary software is safer and cheaper.
Least privilege for agents is not a smaller role assigned once. It is authority narrowed by identity, task, resource, action, time and consequence at every tool call.
Deleting a chat is not the same as removing personal data from memory, retrieval, logs, caches, evaluations, backups and learned models.
Most APIs authenticate well and authorise badly. The defects that matter are object-level and function-level access rules scattered across handlers, where no single reviewer can see the whole rule.
Every post-quantum migration plan begins with an inventory, and most inventories stall because the team looks for algorithms instead of looking for the places algorithms are decided.
A passing patch is only the start. Repository evaluation should test understanding, change quality, security, reviewability and recovery.
Model launches are not transformation milestones. Organize the roadmap around workflow, evidence, authority, operating and expansion decisions.
KEEP EXPLORING
Read the method, assumptions and questions to ask.