Dastan Aitzhanov builds AI systems and writes about delegated authority, seen as an owner: who grants it, where it ends, and who answers for the result.https://dastan.aitzhanov.com/en-ushttps://dastan.aitzhanov.com/writing/govern-the-boundary/https://dastan.aitzhanov.com/writing/govern-the-boundary/Model-based rules assume capability is scarce. As capable models become small and cheap, the durable place to attach policy is the layer that decides what an AI system may do.Sat, 26 Sep 2026 12:00:00 GMTGovernanceAgentsSystemshttps://dastan.aitzhanov.com/ideas/an-agent-can-own-the-mistake/https://dastan.aitzhanov.com/ideas/an-agent-can-own-the-mistake/Accountability means bearing the downside. An agent that spends money it cannot repay needs a limit set before it acts, not an apology after.Sat, 26 Sep 2026 12:00:00 GMTIdeasAgentsGovernanceEconomicshttps://dastan.aitzhanov.com/ideas/treat-ai-adoption-as-a-portfolio/https://dastan.aitzhanov.com/ideas/treat-ai-adoption-as-a-portfolio/Any single AI initiative may fail. The portfolio has to pay. Each one needs a cost, an expected return, an owner, and a moment when someone decides to scale it or stop it.Sat, 26 Sep 2026 12:00:00 GMTIdeasEconomicsGovernanceProducthttps://dastan.aitzhanov.com/ideas/whoever-assembles-the-tools-owns-the-boundary/https://dastan.aitzhanov.com/ideas/whoever-assembles-the-tools-owns-the-boundary/A model developer can say what a model was trained to do. Only the party that connected the tools and set the permissions can say what the system is allowed to do.Sat, 26 Sep 2026 12:00:00 GMTIdeasGovernanceAgentsSystemshttps://dastan.aitzhanov.com/ideas/you-can-only-govern-what-you-can-diff/https://dastan.aitzhanov.com/ideas/you-can-only-govern-what-you-can-diff/A rule that lives in a prompt or a context window changes without anyone deciding it should. Governance needs behavior that can be versioned and compared.Sat, 26 Sep 2026 12:00:00 GMTIdeasGovernanceSystemshttps://dastan.aitzhanov.com/ideas/an-agent-should-have-its-own-identity/https://dastan.aitzhanov.com/ideas/an-agent-should-have-its-own-identity/An agent that acts under a person's borrowed credentials cannot be scoped, revoked, or traced apart from that person. Delegation needs a principal of its own.Sat, 26 Sep 2026 12:00:00 GMTIdeasGovernanceAgentshttps://dastan.aitzhanov.com/writing/capacity-is-not-consent/https://dastan.aitzhanov.com/writing/capacity-is-not-consent/Distributed compute is usually framed as a problem of finding available machines. The harder question is who is allowed to decide what those machines do.Fri, 25 Sep 2026 12:00:00 GMTGovernanceEconomicsSystemshttps://dastan.aitzhanov.com/ideas/autonomy-should-scale-with-reversibility/https://dastan.aitzhanov.com/ideas/autonomy-should-scale-with-reversibility/The useful question is not whether an AI agent is autonomous, but how much autonomy makes sense for a particular action.Fri, 25 Sep 2026 12:00:00 GMTIdeasAgentsGovernancehttps://dastan.aitzhanov.com/ideas/clarification-is-part-of-intelligence/https://dastan.aitzhanov.com/ideas/clarification-is-part-of-intelligence/A system that confidently completes the wrong task is not more intelligent than one that knows when it needs to ask.Fri, 25 Sep 2026 12:00:00 GMTIdeasAgentsProducthttps://dastan.aitzhanov.com/ideas/leasing-can-be-part-of-product-design/https://dastan.aitzhanov.com/ideas/leasing-can-be-part-of-product-design/For infrastructure products, a lease can define control, maintenance, service expectations, and what happens when equipment ages.Fri, 25 Sep 2026 12:00:00 GMTIdeasEconomicsProducthttps://dastan.aitzhanov.com/ideas/the-first-useful-answer-may-be-a-validation-plan/https://dastan.aitzhanov.com/ideas/the-first-useful-answer-may-be-a-validation-plan/Early ideas often need a cheap test of their assumptions more than they need a polished recommendation.Fri, 25 Sep 2026 12:00:00 GMTIdeasProducthttps://dastan.aitzhanov.com/ideas/infrastructure-software-can-become-a-financing-business/https://dastan.aitzhanov.com/ideas/infrastructure-software-can-become-a-financing-business/When software depends on hardware, replacement, insurance, and lifecycle management, the balance sheet becomes part of the product.Fri, 25 Sep 2026 12:00:00 GMTIdeasEconomicsSystemshttps://dastan.aitzhanov.com/ideas/open-source-the-layer-people-need-to-trust/https://dastan.aitzhanov.com/ideas/open-source-the-layer-people-need-to-trust/The boundary between open and proprietary should follow where independent inspection matters, not a general philosophy about software.Fri, 25 Sep 2026 12:00:00 GMTIdeasGovernanceSystemshttps://dastan.aitzhanov.com/ideas/tokens-may-be-the-wrong-unit/https://dastan.aitzhanov.com/ideas/tokens-may-be-the-wrong-unit/Token count is easy to meter, but it does not represent the same work, service quality, or value across systems and tasks.Fri, 25 Sep 2026 12:00:00 GMTIdeasEconomicsSystemshttps://dastan.aitzhanov.com/ideas/a-goal-is-more-useful-than-a-persona/https://dastan.aitzhanov.com/ideas/a-goal-is-more-useful-than-a-persona/A persona shapes how an assistant speaks; a goal tells the system what future state it is trying to help create.Fri, 25 Sep 2026 12:00:00 GMTIdeasProductAgentshttps://dastan.aitzhanov.com/ideas/more-review-is-not-always-more-safety/https://dastan.aitzhanov.com/ideas/more-review-is-not-always-more-safety/Safety should come from clear gates around consequential changes, not from making every small change wait for everything else.Fri, 25 Sep 2026 12:00:00 GMTIdeasGovernanceProducthttps://dastan.aitzhanov.com/ideas/a-model-should-know-which-mode-the-user-is-in/https://dastan.aitzhanov.com/ideas/a-model-should-know-which-mode-the-user-is-in/The same request needs different behavior when a person is exploring, deciding, building, or authorizing action.Fri, 25 Sep 2026 12:00:00 GMTIdeasAgentsProducthttps://dastan.aitzhanov.com/ideas/the-system-should-remember-decisions/https://dastan.aitzhanov.com/ideas/the-system-should-remember-decisions/Durable state is the record of what was chosen, why, and what evidence would cause the choice to change.Fri, 25 Sep 2026 12:00:00 GMTIdeasSystemsProducthttps://dastan.aitzhanov.com/ideas/coding-agents-can-optimize-the-wrong-stage/https://dastan.aitzhanov.com/ideas/coding-agents-can-optimize-the-wrong-stage/An agent can produce excellent implementation work while the product is still asking a more important, earlier question.Fri, 25 Sep 2026 12:00:00 GMTIdeasAgentsProducthttps://dastan.aitzhanov.com/ideas/declarative-intelligence-may-be-the-missing-layer/https://dastan.aitzhanov.com/ideas/declarative-intelligence-may-be-the-missing-layer/Many agent decisions need a constrained label, score, or probability, not another open-ended paragraph.Fri, 25 Sep 2026 12:00:00 GMTIdeasAgentsSystemshttps://dastan.aitzhanov.com/ideas/good-systems-expose-their-assumptions/https://dastan.aitzhanov.com/ideas/good-systems-expose-their-assumptions/A conclusion becomes more useful when the assumptions supporting it, and the evidence that could break it, are visible.Fri, 25 Sep 2026 12:00:00 GMTIdeasProductGovernancehttps://dastan.aitzhanov.com/ideas/software-should-mature-through-four-stages/https://dastan.aitzhanov.com/ideas/software-should-mature-through-four-stages/AI-assisted products move from novelty to usefulness, dependability, and refinement, and each stage needs different work.Fri, 25 Sep 2026 12:00:00 GMTIdeasProductSystemshttps://dastan.aitzhanov.com/ideas/tool-selection-is-part-of-the-decision/https://dastan.aitzhanov.com/ideas/tool-selection-is-part-of-the-decision/Choosing a tool also chooses a permission boundary, an execution path, and a particular risk surface.Fri, 25 Sep 2026 12:00:00 GMTIdeasAgentsGovernancehttps://dastan.aitzhanov.com/ideas/analysis-will-be-abundant-before-judgment/https://dastan.aitzhanov.com/ideas/analysis-will-be-abundant-before-judgment/Generating more comparisons and recommendations does not create ownership of the decision or the judgment to make it well.Fri, 25 Sep 2026 12:00:00 GMTIdeasProductAgentshttps://dastan.aitzhanov.com/ideas/provenance-should-include-where-it-ran/https://dastan.aitzhanov.com/ideas/provenance-should-include-where-it-ran/Model identity is only part of provenance when execution location changes privacy, latency, cost, and trust.Fri, 25 Sep 2026 12:00:00 GMTIdeasGovernanceLocal AIhttps://dastan.aitzhanov.com/ideas/local-ai-is-a-scheduling-problem/https://dastan.aitzhanov.com/ideas/local-ai-is-a-scheduling-problem/The practical limit of local AI is often useful throughput, routing, and graceful fallback rather than headline model size.Fri, 25 Sep 2026 12:00:00 GMTIdeasLocal AISystemshttps://dastan.aitzhanov.com/ideas/parallel-agents-need-a-dependency-graph/https://dastan.aitzhanov.com/ideas/parallel-agents-need-a-dependency-graph/More agents do not automatically make work faster; parallelism becomes useful only when dependencies are explicit.Fri, 25 Sep 2026 12:00:00 GMTIdeasAgentsSystemshttps://dastan.aitzhanov.com/ideas/resource-sharing-moves-the-capital-problem/https://dastan.aitzhanov.com/ideas/resource-sharing-moves-the-capital-problem/Spare compute may look capital-light to a network, but the assets, depreciation, energy, and maintenance still belong to someone.Fri, 25 Sep 2026 12:00:00 GMTIdeasEconomicsLocal AIhttps://dastan.aitzhanov.com/ideas/the-most-important-context-is-the-trajectory/https://dastan.aitzhanov.com/ideas/the-most-important-context-is-the-trajectory/A personal system needs more than facts about the present; it needs to understand the direction the person is trying to move.Fri, 25 Sep 2026 12:00:00 GMTIdeasProductAgentshttps://dastan.aitzhanov.com/ideas/token-efficiency-is-an-architectural-concern/https://dastan.aitzhanov.com/ideas/token-efficiency-is-an-architectural-concern/Repeatedly rediscovering context and decisions is not merely expensive; it is evidence that the system has forgotten its own structure.Fri, 25 Sep 2026 12:00:00 GMTIdeasSystemsEconomicshttps://dastan.aitzhanov.com/ideas/a-model-can-recommend-authority/https://dastan.aitzhanov.com/ideas/a-model-can-recommend-authority/Model judgment can inform a policy, but authority should come from something explicit enough to review and test.Fri, 25 Sep 2026 12:00:00 GMTIdeasGovernanceAgentshttps://dastan.aitzhanov.com/ideas/an-audit-log-is-not-a-control/https://dastan.aitzhanov.com/ideas/an-audit-log-is-not-a-control/A record written after an action is evidence; it cannot substitute for a rule capable of stopping the action beforehand.Fri, 25 Sep 2026 12:00:00 GMTIdeasGovernanceSystemshttps://dastan.aitzhanov.com/ideas/permissions-should-expire-with-the-work/https://dastan.aitzhanov.com/ideas/permissions-should-expire-with-the-work/Agent authority should be narrow, short-lived, and tied to a particular identity, action, resource, and time window.Fri, 25 Sep 2026 12:00:00 GMTIdeasGovernanceAgentshttps://dastan.aitzhanov.com/ideas/the-place-that-performs-the-action-should-say-no/https://dastan.aitzhanov.com/ideas/the-place-that-performs-the-action-should-say-no/Policy becomes real only when the system that sends, deletes, spends, or changes a record can refuse the action.Fri, 25 Sep 2026 12:00:00 GMTIdeasGovernanceSystemshttps://dastan.aitzhanov.com/ideas/a-personal-ai-appliance-should-feel-boring/https://dastan.aitzhanov.com/ideas/a-personal-ai-appliance-should-feel-boring/Personal AI becomes successful infrastructure when availability, privacy, resource use, and recovery stop demanding attention.Fri, 25 Sep 2026 12:00:00 GMTIdeasLocal AIProducthttps://dastan.aitzhanov.com/ideas/consumer-hardware-becomes-infrastructure/https://dastan.aitzhanov.com/ideas/consumer-hardware-becomes-infrastructure/Powerful machines become dependable infrastructure only after variation in hardware, runtimes, networks, and failure modes is absorbed.Fri, 25 Sep 2026 12:00:00 GMTIdeasLocal AISystemshttps://dastan.aitzhanov.com/ideas/the-customer-should-not-need-the-architecture-lesson/https://dastan.aitzhanov.com/ideas/the-customer-should-not-need-the-architecture-lesson/Architecture earns its place when customers can feel the privacy, speed, resilience, control, or lower cost it creates.Fri, 25 Sep 2026 12:00:00 GMTIdeasProducthttps://dastan.aitzhanov.com/ideas/the-frontend-is-the-door/https://dastan.aitzhanov.com/ideas/the-frontend-is-the-door/The interface may change, but the user's context, authority, tools, and work should remain coherent behind it.Fri, 25 Sep 2026 12:00:00 GMTIdeasProductSystemshttps://dastan.aitzhanov.com/ideas/failure-should-reduce-capability/https://dastan.aitzhanov.com/ideas/failure-should-reduce-capability/A resilient AI system should fall back to a smaller, safer set of capabilities when one component becomes unavailable.Fri, 25 Sep 2026 12:00:00 GMTIdeasSystemsLocal AIhttps://dastan.aitzhanov.com/ideas/the-best-escalation-path-is-a-product-feature/https://dastan.aitzhanov.com/ideas/the-best-escalation-path-is-a-product-feature/What happens when an AI system is uncertain often matters more than another improvement to its average-case benchmark.Fri, 25 Sep 2026 12:00:00 GMTIdeasGovernanceProducthttps://dastan.aitzhanov.com/writing/probabilistic-intelligence-deterministic-control/https://dastan.aitzhanov.com/writing/probabilistic-intelligence-deterministic-control/Generative models are for reasoning that is still open. Repeated decisions belong to a narrower model, and repeated actions belong to ordinary software.Mon, 21 Sep 2026 12:00:00 GMTAgentsGovernanceSystemshttps://dastan.aitzhanov.com/ideas/novelty-belongs-to-models/https://dastan.aitzhanov.com/ideas/novelty-belongs-to-models/Use a generative model to find a path the first time. Once the path repeats, call it. Do not pay a model to rediscover it.Mon, 21 Sep 2026 12:00:00 GMTIdeasAgentsSystemshttps://dastan.aitzhanov.com/ideas/policy-should-not-be-probabilistic/https://dastan.aitzhanov.com/ideas/policy-should-not-be-probabilistic/A classifier may return a probability. The rule that permits an action should still be code you can read, version, and test.Mon, 21 Sep 2026 12:00:00 GMTIdeasGovernanceSystemshttps://dastan.aitzhanov.com/ideas/the-best-agent-workflow-stops-looking-agentic/https://dastan.aitzhanov.com/ideas/the-best-agent-workflow-stops-looking-agentic/An agent loop is a reasonable way to discover a procedure. It is a poor way to run one you already trust.Mon, 21 Sep 2026 12:00:00 GMTIdeasAgentsSystemshttps://dastan.aitzhanov.com/writing/what-owners-notice/https://dastan.aitzhanov.com/writing/what-owners-notice/Once analysis is cheap, an organization is still defined by who may commit. A note on ownership, beside the technical thread.Sun, 20 Sep 2026 12:00:00 GMTEconomicsGovernancehttps://dastan.aitzhanov.com/ideas/buy-the-decision/https://dastan.aitzhanov.com/ideas/buy-the-decision/The software worth trusting in a business shortens the path from a fact to a commitment someone will sign.Fri, 18 Sep 2026 12:00:00 GMTIdeasProductEconomicshttps://dastan.aitzhanov.com/ideas/routing-is-a-product-decision/https://dastan.aitzhanov.com/ideas/routing-is-a-product-decision/Choosing a path is not only a cost and latency problem. The policy says what good enough means, and whether the user can see the switch.Wed, 16 Sep 2026 12:00:00 GMTIdeasAgentsProductGovernancehttps://dastan.aitzhanov.com/writing/the-computer-should-be-personal/https://dastan.aitzhanov.com/writing/the-computer-should-be-personal/Local inference matters less as a benchmark and more as a question of where a person's thinking is allowed to live.Mon, 14 Sep 2026 12:00:00 GMTLocal AISystemshttps://dastan.aitzhanov.com/ideas/an-agent-needs-a-job/https://dastan.aitzhanov.com/ideas/an-agent-needs-a-job/A role and a manner of speaking make a demo. A job has inputs, a definition of done, and a person who owns the outcome.Thu, 10 Sep 2026 12:00:00 GMTIdeasAgentsProducthttps://dastan.aitzhanov.com/writing/context-is-the-product/https://dastan.aitzhanov.com/writing/context-is-the-product/What a model is allowed to see is one boundary. What it is allowed to decide, and what the system is allowed to do, are the other two.Tue, 08 Sep 2026 12:00:00 GMTSystemsGovernance