https://www.gunnertech.com/blog Straight answers about autonomous software factories, agent-run delivery and what happens to teams that refuse to adapt. en Sat, 31 Oct 2026 04:00:00 GMT https://www.gunnertech.com/blog/adoption-is-not-an-operating-model https://www.gunnertech.com/blog/adoption-is-not-an-operating-model Sat, 31 Oct 2026 04:00:00 GMT When every software team has AI tools, advantage comes from the factory that governs their work and turns output into accepted change. https://www.gunnertech.com/blog/agent-complaints-are-factory-requirements https://www.gunnertech.com/blog/agent-complaints-are-factory-requirements Sat, 31 Oct 2026 04:00:00 GMT Coding-agent failures become solvable when teams turn each complaint into a routing, authority, proof, or recovery requirement for the factory. https://www.gunnertech.com/blog/repository-search-belongs-in-the-agent-budget https://www.gunnertech.com/blog/repository-search-belongs-in-the-agent-budget Sat, 31 Oct 2026 04:00:00 GMT Repository searches and tool results consume model capacity before coding starts; meter exploration and require evidence before widening the search. https://www.gunnertech.com/blog/security-has-to-run-at-generation-speed https://www.gunnertech.com/blog/security-has-to-run-at-generation-speed Sat, 31 Oct 2026 04:00:00 GMT Agent-written code changes the shape and volume of security risk, so the factory must enforce boundaries and produce proof on every change. https://www.gunnertech.com/blog/your-agents-need-a-learning-loop https://www.gunnertech.com/blog/your-agents-need-a-learning-loop Sat, 31 Oct 2026 04:00:00 GMT Agent-run software delivery improves when production evidence becomes a durable rule for the next planning and build cycle. https://www.gunnertech.com/blog/your-cache-is-part-of-the-model-comparison https://www.gunnertech.com/blog/your-cache-is-part-of-the-model-comparison Sat, 31 Oct 2026 04:00:00 GMT Compare models on accepted work under warm, cold and fallback conditions, including cache writes, retries and the cost of switching. https://www.gunnertech.com/blog/your-productivity-gain-is-waiting-in-a-queue https://www.gunnertech.com/blog/your-productivity-gain-is-waiting-in-a-queue Sat, 31 Oct 2026 04:00:00 GMT AI coding speed becomes business value only when the whole delivery system can absorb, prove, release, and learn from the extra work. https://www.gunnertech.com/blog/authority-has-to-travel-with-the-work https://www.gunnertech.com/blog/authority-has-to-travel-with-the-work Tue, 20 Oct 2026 04:00:00 GMT Agent-run software delivery needs authority attached to each action, enforced where the work happens, and proven after the fact. https://www.gunnertech.com/blog/architecture-has-to-become-agent-context https://www.gunnertech.com/blog/architecture-has-to-become-agent-context Mon, 19 Oct 2026 04:00:00 GMT Coding agents need architecture delivered as executable context, with boundaries, evidence, and consequences attached to the work. https://www.gunnertech.com/blog/a-cheaper-model-is-a-routing-signal https://www.gunnertech.com/blog/a-cheaper-model-is-a-routing-signal Mon, 05 Oct 2026 04:00:00 GMT Model price changes should alter factory routing only after task-level evidence proves where cheaper execution holds and where it fails. https://www.gunnertech.com/blog/a-factory-needs-a-goal-loop https://www.gunnertech.com/blog/a-factory-needs-a-goal-loop Mon, 05 Oct 2026 04:00:00 GMT A software factory needs a measurable goal, durable work state, and a loop that keeps checking reality after the release. https://www.gunnertech.com/blog/a-launch-day-is-not-a-factory-strategy https://www.gunnertech.com/blog/a-launch-day-is-not-a-factory-strategy Mon, 05 Oct 2026 04:00:00 GMT New models and agent platforms matter only after your factory can route, govern, measure, and replace them without rebuilding the operation. https://www.gunnertech.com/blog/a-skills-gap-is-a-system-gap https://www.gunnertech.com/blog/a-skills-gap-is-a-system-gap Mon, 05 Oct 2026 04:00:00 GMT Projects stall when critical knowledge lives in scarce people instead of the specifications, controls, and agent routes that can carry the work. https://www.gunnertech.com/blog/ai-native-needs-stable-infrastructure https://www.gunnertech.com/blog/ai-native-needs-stable-infrastructure Mon, 05 Oct 2026 04:00:00 GMT AI-native delivery starts with evidence, deployment, observation, and recovery systems that can safely carry more autonomous work. https://www.gunnertech.com/blog/ai-subscriptions-are-becoming-capacity-contracts https://www.gunnertech.com/blog/ai-subscriptions-are-becoming-capacity-contracts Mon, 05 Oct 2026 04:00:00 GMT Model subscriptions are changing into capacity purchases, so software factories need to route work by outcome, speed, and evidence instead of seats. https://www.gunnertech.com/blog/an-explanation-is-not-a-corrective-action https://www.gunnertech.com/blog/an-explanation-is-not-a-corrective-action Mon, 05 Oct 2026 04:00:00 GMT A useful incident review changes the factory's gates, routes, permissions, or evidence instead of asking people to remember the lesson. https://www.gunnertech.com/blog/build-the-factory-backward-from-the-verdict https://www.gunnertech.com/blog/build-the-factory-backward-from-the-verdict Mon, 05 Oct 2026 04:00:00 GMT Start an agent-run software factory with independent release evidence, then build the specification and implementation path that can satisfy it. https://www.gunnertech.com/blog/cheaper-intelligence-raises-the-factory-bar https://www.gunnertech.com/blog/cheaper-intelligence-raises-the-factory-bar Mon, 05 Oct 2026 04:00:00 GMT As capable models get cheaper, the advantage moves to factories that can turn low-cost intelligence into governed, reusable delivery. https://www.gunnertech.com/blog/code-review-needs-a-named-job https://www.gunnertech.com/blog/code-review-needs-a-named-job Mon, 05 Oct 2026 04:00:00 GMT Code review works when each stop names the consequence, question, and authority involved instead of asking one reviewer to carry the whole delivery system. https://www.gunnertech.com/blog/coding-is-not-the-product https://www.gunnertech.com/blog/coding-is-not-the-product Mon, 05 Oct 2026 04:00:00 GMT Cheaper code changes software economics, but ownership only scales when requirements, proof, and accountability move into the factory. https://www.gunnertech.com/blog/context-engineering-is-factory-design https://www.gunnertech.com/blog/context-engineering-is-factory-design Mon, 05 Oct 2026 04:00:00 GMT Reliable coding agents need a governed context route that selects current rules, task state, evidence, and tool feedback for each job. https://www.gunnertech.com/blog/minimalism-needs-an-extension-boundary https://www.gunnertech.com/blog/minimalism-needs-an-extension-boundary Mon, 05 Oct 2026 04:00:00 GMT A minimal agent harness stays dependable when experiments cross an explicit extension boundary instead of quietly enlarging the core. https://www.gunnertech.com/blog/model-choice-belongs-after-the-gates https://www.gunnertech.com/blog/model-choice-belongs-after-the-gates Mon, 05 Oct 2026 04:00:00 GMT Route agent work by consequence, evidence, and review coverage before paying a premium for a model that fails differently. https://www.gunnertech.com/blog/popularity-does-not-pay-the-maintainers https://www.gunnertech.com/blog/popularity-does-not-pay-the-maintainers Mon, 05 Oct 2026 04:00:00 GMT A critical open-source dependency needs an ownership, funding, upgrade, and exit plan before its maintainer's business becomes your emergency. https://www.gunnertech.com/blog/product-sense-has-to-leave-your-head https://www.gunnertech.com/blog/product-sense-has-to-leave-your-head Mon, 05 Oct 2026 04:00:00 GMT Agent-run delivery needs product judgment encoded in specifications, boundaries, evidence, and feedback instead of trapped in one person's intuition. https://www.gunnertech.com/blog/protected-time-is-not-an-operating-model https://www.gunnertech.com/blog/protected-time-is-not-an-operating-model Mon, 05 Oct 2026 04:00:00 GMT AI transformation needs owned capacity, measurable outcomes, and factory controls—not spare hours that ordinary work can quietly reclaim. https://www.gunnertech.com/blog/quality-needs-a-closed-loop https://www.gunnertech.com/blog/quality-needs-a-closed-loop Mon, 05 Oct 2026 04:00:00 GMT More tests and reviews only protect agent-run delivery when failures change the specification, gates, routing, or permissions for the next run. https://www.gunnertech.com/blog/the-decision-boundary-is-the-real-automation-boundary https://www.gunnertech.com/blog/the-decision-boundary-is-the-real-automation-boundary Mon, 05 Oct 2026 04:00:00 GMT Agents can execute most of a workflow without owning the decisions that define success, risk, and permission. https://www.gunnertech.com/blog/the-engineer-becomes-the-cross-examiner https://www.gunnertech.com/blog/the-engineer-becomes-the-cross-examiner Mon, 05 Oct 2026 04:00:00 GMT As agents take over implementation, engineers move toward defining claims, demanding evidence, and turning disagreements into factory controls. https://www.gunnertech.com/blog/the-feature-list-is-not-the-factory https://www.gunnertech.com/blog/the-feature-list-is-not-the-factory Mon, 05 Oct 2026 04:00:00 GMT Evaluate a software factory by the decisions and evidence it connects, not by the number of agents, dashboards, or collaboration surfaces it displays. https://www.gunnertech.com/blog/the-plugin-is-part-of-the-factory https://www.gunnertech.com/blog/the-plugin-is-part-of-the-factory Mon, 05 Oct 2026 04:00:00 GMT Composable agent plugins turn tools, interfaces, schedules, and teams into factory components, so each extension needs an operating contract. https://www.gunnertech.com/blog/your-agent-subscription-is-a-temporary-discount https://www.gunnertech.com/blog/your-agent-subscription-is-a-temporary-discount Mon, 05 Oct 2026 04:00:00 GMT Flat agent subscriptions hide the cost of each accepted outcome, so software factories need usage budgets, routing, and proof before pricing changes. https://www.gunnertech.com/blog/your-agents-need-an-admission-gate https://www.gunnertech.com/blog/your-agents-need-an-admission-gate Mon, 05 Oct 2026 04:00:00 GMT Parallel coding agents only improve delivery when the factory limits work in progress and admits changes at the rate it can prove them. https://www.gunnertech.com/blog/code-is-becoming-a-factory-artifact https://www.gunnertech.com/blog/code-is-becoming-a-factory-artifact Sat, 03 Oct 2026 04:00:00 GMT When agents write most of the code, the durable engineering work moves into specifications, abstractions, gates, and production evidence. https://www.gunnertech.com/blog/the-finish-line-is-a-factory-interface https://www.gunnertech.com/blog/the-finish-line-is-a-factory-interface Sat, 03 Oct 2026 04:00:00 GMT Long-running agents become useful when done, blocked, and dangerous are explicit states the factory can test and route. https://www.gunnertech.com/blog/the-harness-is-the-operating-system https://www.gunnertech.com/blog/the-harness-is-the-operating-system Sat, 03 Oct 2026 04:00:00 GMT An agent harness turns a capable model into governed production machinery by controlling its context, tools, permissions, evidence, and recovery loop. https://www.gunnertech.com/blog/ownership-starts-when-the-agent-stops https://www.gunnertech.com/blog/ownership-starts-when-the-agent-stops Fri, 02 Oct 2026 04:00:00 GMT Business teams can direct agents to build software, but real ownership includes operating, repairing, governing, and eventually retiring what gets built. https://www.gunnertech.com/blog/the-spec-has-to-close-the-loop https://www.gunnertech.com/blog/the-spec-has-to-close-the-loop Thu, 01 Oct 2026 04:00:00 GMT A specification becomes factory machinery only when implementation, proof, and production evidence can change the next run. https://www.gunnertech.com/blog/the-workflow-has-to-survive-the-agent https://www.gunnertech.com/blog/the-workflow-has-to-survive-the-agent Wed, 30 Sep 2026 04:00:00 GMT Coding agents become dependable when plans, state, evidence, and recovery live in the workflow instead of one model session. https://www.gunnertech.com/blog/rework-belongs-in-the-productivity-number https://www.gunnertech.com/blog/rework-belongs-in-the-productivity-number Mon, 28 Sep 2026 04:00:00 GMT Agent productivity becomes measurable when the factory subtracts repair, rescue, and repeated mistakes from accepted outcomes. https://www.gunnertech.com/blog/cheap-judgment-still-has-to-earn-authority https://www.gunnertech.com/blog/cheap-judgment-still-has-to-earn-authority Sat, 19 Sep 2026 04:00:00 GMT A low-cost decision model becomes useful when the factory lets it observe first, measures its misses, and expands its authority one consequence at a time. https://www.gunnertech.com/blog/confidence-needs-a-cost-function https://www.gunnertech.com/blog/confidence-needs-a-cost-function Sat, 19 Sep 2026 04:00:00 GMT A calibrated routing score becomes useful only when the factory prices mistakes, detects drift, and controls what each decision may trigger. https://www.gunnertech.com/blog/the-router-is-not-the-factory https://www.gunnertech.com/blog/the-router-is-not-the-factory Sat, 19 Sep 2026 04:00:00 GMT Small decision models can make agent workflows cheaper and faster, but routing only becomes valuable inside a factory that can prove the decisions hold. https://www.gunnertech.com/blog/a-proof-cannot-rescue-the-wrong-problem https://www.gunnertech.com/blog/a-proof-cannot-rescue-the-wrong-problem Fri, 18 Sep 2026 04:00:00 GMT Agent-run software factories need a research gate before implementation because perfect proof can still certify the wrong specification. https://www.gunnertech.com/blog/the-ide-is-a-manual-control-loop https://www.gunnertech.com/blog/the-ide-is-a-manual-control-loop Fri, 18 Sep 2026 04:00:00 GMT Coding agents become factory assets when repeatable work can leave the editor and run through governed, observable routes. https://www.gunnertech.com/blog/your-ai-champion-is-a-bootstrap-phase https://www.gunnertech.com/blog/your-ai-champion-is-a-bootstrap-phase Fri, 18 Sep 2026 04:00:00 GMT An internal AI champion can start adoption, but the factory has to turn that person's discoveries into shared, enforced operating machinery. https://www.gunnertech.com/blog/autonomy-stops-where-observation-stops https://www.gunnertech.com/blog/autonomy-stops-where-observation-stops Thu, 17 Sep 2026 04:00:00 GMT A self-driving codebase needs an agent-visible route from failure to reproduction, repair, proof, and durable factory improvement. https://www.gunnertech.com/blog/governance-has-to-run-at-agent-speed https://www.gunnertech.com/blog/governance-has-to-run-at-agent-speed Thu, 17 Sep 2026 04:00:00 GMT Agent-run delivery needs executable controls that classify risk, constrain authority, and preserve proof without waiting for a meeting. https://www.gunnertech.com/blog/quality-is-part-of-the-agent-price https://www.gunnertech.com/blog/quality-is-part-of-the-agent-price Thu, 17 Sep 2026 04:00:00 GMT Coding-agent costs only become comparable after the factory counts verification, rework, human rescue, and the changes that actually survive. https://www.gunnertech.com/blog/you-are-benchmarking-the-harness https://www.gunnertech.com/blog/you-are-benchmarking-the-harness Thu, 17 Sep 2026 04:00:00 GMT Coding-agent results belong to the whole operating harness, so model selection has to test the route that will actually ship software. https://www.gunnertech.com/blog/trust-should-be-paid-for-once https://www.gunnertech.com/blog/trust-should-be-paid-for-once Wed, 16 Sep 2026 04:00:00 GMT Agent factories waste time and tokens when every session has to rediscover its role, authority, evidence, and recovery path. https://www.gunnertech.com/blog/repetition-is-the-factory-test https://www.gunnertech.com/blog/repetition-is-the-factory-test Tue, 15 Sep 2026 04:00:00 GMT A software factory earns its name when intent, state, proof, and recovery survive the first successful agent run. https://www.gunnertech.com/blog/the-agent-is-not-where-your-product-lives https://www.gunnertech.com/blog/the-agent-is-not-where-your-product-lives Tue, 15 Sep 2026 04:00:00 GMT Durable agent-run delivery comes from machinery that assembles context, controls tools, tests outcomes, and survives model changes. https://www.gunnertech.com/blog/every-feature-needs-an-exit https://www.gunnertech.com/blog/every-feature-needs-an-exit Mon, 14 Sep 2026 04:00:00 GMT Cheap code makes features easy to add, so software factories must define observation, removal triggers, and cleanup before anything ships. https://www.gunnertech.com/blog/the-harness-is-a-security-boundary https://www.gunnertech.com/blog/the-harness-is-a-security-boundary Mon, 14 Sep 2026 04:00:00 GMT Agent security fails when organizations vet the model but leave its tools, memory, retries, permissions, and proof outside the boundary. https://www.gunnertech.com/blog/a-second-agent-is-not-independent-proof https://www.gunnertech.com/blog/a-second-agent-is-not-independent-proof Sat, 12 Sep 2026 04:00:00 GMT Agent-run testing becomes trustworthy only when the verifier has separate authority, preserved evidence, and no power to redefine success. https://www.gunnertech.com/blog/ai-code-needs-a-higher-bar https://www.gunnertech.com/blog/ai-code-needs-a-higher-bar Sat, 12 Sep 2026 04:00:00 GMT Agent-written production code needs risk-based gates that match its speed, reach, and ability to repeat the same mistake at scale. https://www.gunnertech.com/blog/coding-is-a-hobby-delivery-is-the-job https://www.gunnertech.com/blog/coding-is-a-hobby-delivery-is-the-job Sat, 12 Sep 2026 04:00:00 GMT Agents are making hand-written code optional, but reliable software still requires judgment, governed delivery, and proof that survives production. https://www.gunnertech.com/blog/creativity-is-not-a-release-decision https://www.gunnertech.com/blog/creativity-is-not-a-release-decision Sat, 12 Sep 2026 04:00:00 GMT Generative AI makes plausible ideas abundant, so software factories must move human judgment into selection, consequences, and proof. https://www.gunnertech.com/blog/recursive-improvement-is-a-factory-problem https://www.gunnertech.com/blog/recursive-improvement-is-a-factory-problem Sat, 12 Sep 2026 04:00:00 GMT AI research can accelerate itself only when experiments, evidence, authority, and rollback form a governed operating loop. https://www.gunnertech.com/blog/the-factory-is-built-from-refusals https://www.gunnertech.com/blog/the-factory-is-built-from-refusals Sat, 12 Sep 2026 04:00:00 GMT A real software factory controls agent output with explicit intake, isolation, verification, and release gates that can stop the work. https://www.gunnertech.com/blog/a-green-gate-can-still-miss-the-change https://www.gunnertech.com/blog/a-green-gate-can-still-miss-the-change Fri, 11 Sep 2026 04:00:00 GMT A validation suite is only trustworthy when it proves that every changed surface reached the checks that claim authority over it. https://www.gunnertech.com/blog/a-model-launch-is-not-a-capacity-plan https://www.gunnertech.com/blog/a-model-launch-is-not-a-capacity-plan Fri, 11 Sep 2026 04:00:00 GMT A new coding model creates useful competition, but only a governed factory can turn changing capability, cost, and performance into dependable software delivery. https://www.gunnertech.com/blog/disposable-code-needs-a-hard-boundary https://www.gunnertech.com/blog/disposable-code-needs-a-hard-boundary Fri, 11 Sep 2026 04:00:00 GMT Agent factories need separate rules for temporary tool work and durable product code, with gates that stop local shortcuts from becoming permanent maintenance costs. https://www.gunnertech.com/blog/green-tests-are-not-a-clean-bill-of-health https://www.gunnertech.com/blog/green-tests-are-not-a-clean-bill-of-health Fri, 11 Sep 2026 04:00:00 GMT Agent-run delivery needs structural quality gates because correct behavior can still hide duplication, complexity, and expensive future changes. https://www.gunnertech.com/blog/a-testing-tool-is-not-a-testing-strategy https://www.gunnertech.com/blog/a-testing-tool-is-not-a-testing-strategy Tue, 08 Sep 2026 04:00:00 GMT Agents need risk-directed verification, independent oracles, and tests that attack the implementation instead of decorating it. https://www.gunnertech.com/blog/bad-code-does-not-find-the-bottom https://www.gunnertech.com/blog/bad-code-does-not-find-the-bottom Mon, 07 Sep 2026 04:00:00 GMT Software quality has no natural floor, so agent-run factories need explicit stop conditions that reject compounding defects before they become the system. https://www.gunnertech.com/blog/build-buy-or-blend-is-a-routing-decision https://www.gunnertech.com/blog/build-buy-or-blend-is-a-routing-decision Mon, 07 Sep 2026 04:00:00 GMT AI sourcing decisions should route each capability by strategic value, production burden, and the evidence required to keep owning it. https://www.gunnertech.com/blog/rollback-is-a-factory-capability https://www.gunnertech.com/blog/rollback-is-a-factory-capability Mon, 07 Sep 2026 04:00:00 GMT Agent-run delivery needs a tested route out of production, because faster change without mechanical reversal turns every release into a larger bet. https://www.gunnertech.com/blog/the-agent-toolbox-is-a-build-input https://www.gunnertech.com/blog/the-agent-toolbox-is-a-build-input Mon, 07 Sep 2026 04:00:00 GMT When coding agents choose their own tools, the factory needs to govern those choices as versioned, testable build inputs instead of disposable setup. https://www.gunnertech.com/blog/the-demo-is-not-the-operating-model https://www.gunnertech.com/blog/the-demo-is-not-the-operating-model Mon, 07 Sep 2026 04:00:00 GMT AI makes software easier to build, but production ownership still decides whether that software creates an advantage or a recurring liability. https://www.gunnertech.com/blog/you-cannot-retrofit-a-factory-around-a-pull-request https://www.gunnertech.com/blog/you-cannot-retrofit-a-factory-around-a-pull-request Mon, 07 Sep 2026 04:00:00 GMT Agents inside a human delivery pipeline create faster inventory; a software factory redesigns the route so independent evidence can move routine work without human approval. https://www.gunnertech.com/blog/self-hosted-execution-is-not-operational-control https://www.gunnertech.com/blog/self-hosted-execution-is-not-operational-control Thu, 03 Sep 2026 04:00:00 GMT Running agent tools on your own infrastructure improves custody, but control still depends on who plans the work and whether the factory can survive their exit. https://www.gunnertech.com/blog/the-factory-needs-a-complexity-budget https://www.gunnertech.com/blog/the-factory-needs-a-complexity-budget Thu, 03 Sep 2026 04:00:00 GMT Agent-run delivery needs a mechanical complexity budget because a worker that can navigate tangled code will not naturally stop creating more of it. https://www.gunnertech.com/blog/understanding-cannot-wait-for-code-review https://www.gunnertech.com/blog/understanding-cannot-wait-for-code-review Thu, 03 Sep 2026 04:00:00 GMT Agent-run delivery keeps people oriented by moving design judgment, executable constraints, and operating responsibility ahead of the finished diff. https://www.gunnertech.com/blog/the-agent-needs-a-real-system-to-test https://www.gunnertech.com/blog/the-agent-needs-a-real-system-to-test Mon, 31 Aug 2026 04:00:00 GMT Agent verification becomes trustworthy when code runs against production-shaped dependencies without sharing another worker's state. https://www.gunnertech.com/blog/your-ai-vendor-is-a-control-plane https://www.gunnertech.com/blog/your-ai-vendor-is-a-control-plane Sun, 30 Aug 2026 04:00:00 GMT Agent factories stay operable when vendor access is treated as a revocable dependency instead of permanent infrastructure. https://www.gunnertech.com/blog/a-rule-should-get-harder https://www.gunnertech.com/blog/a-rule-should-get-harder Thu, 27 Aug 2026 04:00:00 GMT Agent-run delivery gets safer when recurring judgment moves from written guidance to review gates and finally into deterministic enforcement. https://www.gunnertech.com/blog/cheap-code-makes-complexity-more-expensive https://www.gunnertech.com/blog/cheap-code-makes-complexity-more-expensive Thu, 27 Aug 2026 04:00:00 GMT Agents make code abundant, so software factories have to control dependencies, evidence, and removal before complexity consumes the gain. https://www.gunnertech.com/blog/programming-ends-where-the-factory-begins https://www.gunnertech.com/blog/programming-ends-where-the-factory-begins Thu, 27 Aug 2026 04:00:00 GMT When agents write more code than people can review, architecture, evidence, and operating authority become the real engineering work. https://www.gunnertech.com/blog/the-harness-has-to-learn https://www.gunnertech.com/blog/the-harness-has-to-learn Thu, 27 Aug 2026 04:00:00 GMT An agent harness becomes durable infrastructure when corrections change its context, roles, gates, and routes instead of disappearing after each run. https://www.gunnertech.com/blog/the-product-record-must-survive-the-code https://www.gunnertech.com/blog/the-product-record-must-survive-the-code Thu, 27 Aug 2026 04:00:00 GMT When agents make implementation cheap, product delivery has to preserve intent, behavioral evidence, and runtime consequences as one verified record. https://www.gunnertech.com/blog/discovery-is-not-delegation https://www.gunnertech.com/blog/discovery-is-not-delegation Mon, 24 Aug 2026 04:00:00 GMT Agent registries solve discovery, but a software factory still needs policy, routing, proof, and revocation before an agent can act. https://www.gunnertech.com/blog/expertise-must-move-into-the-factory https://www.gunnertech.com/blog/expertise-must-move-into-the-factory Mon, 24 Aug 2026 04:00:00 GMT AI changes how software expertise forms: routine coding shrinks while judgment, failure analysis, and factory controls become the real training ground. https://www.gunnertech.com/blog/your-voice-needs-a-gate https://www.gunnertech.com/blog/your-voice-needs-a-gate Mon, 24 Aug 2026 04:00:00 GMT AI writing stays recognizable when voice becomes an executable specification with checks, examples, and a feedback loop. https://www.gunnertech.com/blog/a-pattern-is-not-a-factory-control https://www.gunnertech.com/blog/a-pattern-is-not-a-factory-control Sun, 23 Aug 2026 04:00:00 GMT Agentic engineering patterns become valuable when a software factory can apply, verify, and improve them without depending on human memory. https://www.gunnertech.com/blog/the-expensive-model-belongs-at-the-narrowest-gate https://www.gunnertech.com/blog/the-expensive-model-belongs-at-the-narrowest-gate Sun, 23 Aug 2026 04:00:00 GMT Model routing turns cost, capability, context, and data policy into factory controls instead of per-task guesses. https://www.gunnertech.com/blog/the-harness-is-not-the-factory https://www.gunnertech.com/blog/the-harness-is-not-the-factory Sun, 23 Aug 2026 04:00:00 GMT An agent harness gives a model instructions, tools, a working loop, and model choice, but production delivery still needs a factory around it. https://www.gunnertech.com/blog/your-toolchain-is-not-your-operating-system https://www.gunnertech.com/blog/your-toolchain-is-not-your-operating-system Sun, 23 Aug 2026 04:00:00 GMT Fast-changing AI tools stop exhausting teams when models become replaceable parts and verification becomes factory machinery. https://www.gunnertech.com/blog/code-review-cannot-carry-the-factory https://www.gunnertech.com/blog/code-review-cannot-carry-the-factory Sat, 22 Aug 2026 04:00:00 GMT Agent-run delivery needs executable acceptance, independent proof, and controlled release because reading every changed line cannot scale with machine output. https://www.gunnertech.com/blog/coding-speed-is-not-factory-throughput https://www.gunnertech.com/blog/coding-speed-is-not-factory-throughput Sat, 22 Aug 2026 04:00:00 GMT Faster code production matters only when the delivery system can verify, release, and learn from the change without moving the cost downstream. https://www.gunnertech.com/blog/done-is-a-factory-decision https://www.gunnertech.com/blog/done-is-a-factory-decision Sat, 22 Aug 2026 04:00:00 GMT Agent loops need an external definition of progress, acceptance, and stopping before more attempts become expensive motion. https://www.gunnertech.com/blog/a-citizen-developer-is-not-a-software-factory https://www.gunnertech.com/blog/a-citizen-developer-is-not-a-software-factory Thu, 20 Aug 2026 04:00:00 GMT Giving more people an AI build interface only works when a governed factory turns intent into software that can survive production. https://www.gunnertech.com/blog/a-closed-pull-request-is-not-finished-work https://www.gunnertech.com/blog/a-closed-pull-request-is-not-finished-work Thu, 20 Aug 2026 04:00:00 GMT Software factories should measure work from accepted change through production survival, then use that evidence to control routing, authority, and spend. https://www.gunnertech.com/blog/ai-authorship-is-not-a-quality-signal https://www.gunnertech.com/blog/ai-authorship-is-not-a-quality-signal Thu, 20 Aug 2026 04:00:00 GMT The spread of AI-written web pages makes authorship less useful as a shortcut and raises the value of provenance, proof, and accountable publishing systems. https://www.gunnertech.com/blog/junior-engineers-need-a-different-ramp https://www.gunnertech.com/blog/junior-engineers-need-a-different-ramp Thu, 20 Aug 2026 04:00:00 GMT AI expands what junior engineers can attempt, but factories must turn that reach into evidence, feedback, and durable judgment. https://www.gunnertech.com/blog/ai-usage-is-not-factory-performance https://www.gunnertech.com/blog/ai-usage-is-not-factory-performance Wed, 19 Aug 2026 04:00:00 GMT AI adoption data becomes useful only when teams connect agent activity to accepted work, production evidence, cost, and durable factory controls. https://www.gunnertech.com/blog/skepticism-belongs-in-the-gate https://www.gunnertech.com/blog/skepticism-belongs-in-the-gate Mon, 17 Aug 2026 04:00:00 GMT The strongest objections to AI-generated software should become executable production gates, not reasons to preserve manual delivery. https://www.gunnertech.com/blog/the-proof-must-outrank-the-builder https://www.gunnertech.com/blog/the-proof-must-outrank-the-builder Mon, 17 Aug 2026 04:00:00 GMT Agent-run delivery needs executable specifications that can reject a builder's work without letting the builder rewrite the definition of correct. https://www.gunnertech.com/blog/a-bigger-model-does-not-shrink-the-factory https://www.gunnertech.com/blog/a-bigger-model-does-not-shrink-the-factory Sun, 16 Aug 2026 04:00:00 GMT Sparse model architecture can change inference economics, but software delivery still depends on routing, proof, permissions, and production feedback. https://www.gunnertech.com/blog/a-code-graph-is-not-a-factory https://www.gunnertech.com/blog/a-code-graph-is-not-a-factory Sun, 16 Aug 2026 04:00:00 GMT A code graph can route agents through a large repository, but only a governed factory can decide what context matters and prove the change is safe. https://www.gunnertech.com/blog/a-coding-session-is-not-factory-memory https://www.gunnertech.com/blog/a-coding-session-is-not-factory-memory Sun, 16 Aug 2026 04:00:00 GMT A productive coding session can carry useful context, but a software factory must preserve state, proof, and recovery outside the conversation. https://www.gunnertech.com/blog/agility-needs-an-executable-specification https://www.gunnertech.com/blog/agility-needs-an-executable-specification Sun, 16 Aug 2026 04:00:00 GMT Agent-run software delivery needs specifications that turn intent, constraints, and change into an executable production boundary. https://www.gunnertech.com/blog/craft-must-become-machinery https://www.gunnertech.com/blog/craft-must-become-machinery Sun, 16 Aug 2026 04:00:00 GMT AI coding becomes a durable production system when expert judgment turns into shared context, controls, proof, and feedback. https://www.gunnertech.com/blog/history-needs-a-reproduction-test https://www.gunnertech.com/blog/history-needs-a-reproduction-test Sun, 16 Aug 2026 04:00:00 GMT Software factories learn from history by preserving conditions, decisions, evidence, and outcomes as controls that can be tested again. https://www.gunnertech.com/blog/leadership-is-a-manual-control-loop https://www.gunnertech.com/blog/leadership-is-a-manual-control-loop Sun, 16 Aug 2026 04:00:00 GMT Working with AI feels like leadership because humans still carry intent, delegation, review, and correction by hand; a software factory turns that loop into machinery. https://www.gunnertech.com/blog/memory-is-not-a-message-thread https://www.gunnertech.com/blog/memory-is-not-a-message-thread Sun, 16 Aug 2026 04:00:00 GMT Agent factories need governed, retrievable state instead of treating a model's conversation history as durable operational memory. https://www.gunnertech.com/blog/one-score-cannot-route-a-factory https://www.gunnertech.com/blog/one-score-cannot-route-a-factory Sun, 16 Aug 2026 04:00:00 GMT Model scores become useful in production when factories turn capability, cost, authority, and failure evidence into task-specific routing rules. https://www.gunnertech.com/blog/the-rejected-pull-request-trains-the-factory https://www.gunnertech.com/blog/the-rejected-pull-request-trains-the-factory Sun, 16 Aug 2026 04:00:00 GMT Daily agent maintenance becomes dependable when every accepted and rejected change updates the routine that produced it. https://www.gunnertech.com/blog/token-efficiency-is-not-factory-efficiency https://www.gunnertech.com/blog/token-efficiency-is-not-factory-efficiency Sun, 16 Aug 2026 04:00:00 GMT Efficient models cut production costs only when the factory measures accepted work, including retries, proof, and recovery. https://www.gunnertech.com/blog/tokens-need-work-not-users https://www.gunnertech.com/blog/tokens-need-work-not-users Sun, 16 Aug 2026 04:00:00 GMT Durable AI demand will come from governed factories completing consequential work, not from persuading more employees to chat with optional assistants. https://www.gunnertech.com/blog/your-model-contract-is-a-decaying-asset https://www.gunnertech.com/blog/your-model-contract-is-a-decaying-asset Sun, 16 Aug 2026 04:00:00 GMT Inference prices can fall while committed compute costs stay fixed, so software factories must route work by capability, proof, and current cost. https://www.gunnertech.com/blog/your-prompt-is-a-manual-build-step https://www.gunnertech.com/blog/your-prompt-is-a-manual-build-step Sun, 16 Aug 2026 04:00:00 GMT Agent-run delivery scales when the factory assembles task-specific context, routes models, and turns failures into durable controls. https://www.gunnertech.com/blog/code-generation-is-not-the-bottleneck https://www.gunnertech.com/blog/code-generation-is-not-the-bottleneck Thu, 13 Aug 2026 04:00:00 GMT Coding agents move implementation faster, but software factories win by making judgment, proof, and recovery move with it. https://www.gunnertech.com/blog/the-assistant-market-is-a-shrinking-market https://www.gunnertech.com/blog/the-assistant-market-is-a-shrinking-market Thu, 13 Aug 2026 04:00:00 GMT AI market share will move toward agents that complete governed work, not assistants built around jobs the machinery is replacing. https://www.gunnertech.com/blog/the-factory-does-not-need-a-standup https://www.gunnertech.com/blog/the-factory-does-not-need-a-standup Thu, 13 Aug 2026 04:00:00 GMT Agent-run software factories replace Agile ceremony with encoded state, automatic routing, mechanical gates, and durable evidence. https://www.gunnertech.com/blog/the-prompt-box-is-not-the-factory https://www.gunnertech.com/blog/the-prompt-box-is-not-the-factory Thu, 13 Aug 2026 04:00:00 GMT Natural-language software creation expands who can request software, but dependable delivery still requires governed planning, proof, release, and recovery. https://www.gunnertech.com/blog/technical-debt-is-a-judgment-queue https://www.gunnertech.com/blog/technical-debt-is-a-judgment-queue Wed, 12 Aug 2026 04:00:00 GMT Coding agents can execute migrations quickly, but technical debt only falls when the factory recovers the decisions buried inside the old system. https://www.gunnertech.com/blog/a-source-of-truth-needs-a-write-path https://www.gunnertech.com/blog/a-source-of-truth-needs-a-write-path Tue, 11 Aug 2026 04:00:00 GMT Agent-run software delivery needs one governed operational record whose writes, conflicts, evidence, and consequences are mechanically controlled. https://www.gunnertech.com/blog/a-computer-use-score-is-not-operating-authority https://www.gunnertech.com/blog/a-computer-use-score-is-not-operating-authority Mon, 10 Aug 2026 04:00:00 GMT Computer-use benchmarks measure whether a model can act, while a production factory must control what it may change, prove, and recover. https://www.gunnertech.com/blog/four-models-do-not-make-a-factory https://www.gunnertech.com/blog/four-models-do-not-make-a-factory Mon, 10 Aug 2026 04:00:00 GMT Multi-model orchestration creates leverage only when bounded roles, controlled context, and independent gates make coordination cheaper than repetition. https://www.gunnertech.com/blog/the-agent-budget-will-cut-payroll-first https://www.gunnertech.com/blog/the-agent-budget-will-cut-payroll-first Mon, 10 Aug 2026 04:00:00 GMT When agent costs rise, companies will cut repeatable human labor before they abandon machinery that makes software delivery measurable and scalable. https://www.gunnertech.com/blog/your-hours-do-not-compound https://www.gunnertech.com/blog/your-hours-do-not-compound Mon, 10 Aug 2026 04:00:00 GMT Extraordinary products demand extraordinary commitment, but software factories turn founder intensity into durable machinery instead of permanent overtime. https://www.gunnertech.com/blog/auto-permission-is-still-permission https://www.gunnertech.com/blog/auto-permission-is-still-permission Sun, 09 Aug 2026 04:00:00 GMT Coding agents can automate routine permission decisions, but the factory must still bound authority, preserve evidence, and stop unsafe actions mechanically. https://www.gunnertech.com/blog/solved-code-still-needs-a-factory https://www.gunnertech.com/blog/solved-code-still-needs-a-factory Sun, 09 Aug 2026 04:00:00 GMT Coding agents turn implementation into abundant capacity, but only a governed software factory can turn that capacity into reliable business outcomes. https://www.gunnertech.com/blog/the-revenue-is-upstream-because-the-work-is-not https://www.gunnertech.com/blog/the-revenue-is-upstream-because-the-work-is-not Sun, 09 Aug 2026 04:00:00 GMT AI revenue concentrates at model providers while most companies still buy capability instead of operating software factories that capture outcomes. https://www.gunnertech.com/blog/the-workspace-is-becoming-a-control-room https://www.gunnertech.com/blog/the-workspace-is-becoming-a-control-room Sun, 09 Aug 2026 04:00:00 GMT Persistent coding agents turn the development workspace into a factory control room, making authority, routing, evidence, and recovery the real product. https://www.gunnertech.com/blog/trash-output-is-a-factory-signal https://www.gunnertech.com/blog/trash-output-is-a-factory-signal Sun, 09 Aug 2026 04:00:00 GMT Bad AI-generated code is evidence about the delivery system that selected, constrained, tested, and accepted it. https://www.gunnertech.com/blog/cheap-code-does-not-erase-the-craft https://www.gunnertech.com/blog/cheap-code-does-not-erase-the-craft Sat, 08 Aug 2026 04:00:00 GMT Agent-run factories make implementation cheaper without pretending that programming was easy or that engineering judgment no longer matters. https://www.gunnertech.com/blog/headcount-is-not-a-unit-of-work https://www.gunnertech.com/blog/headcount-is-not-a-unit-of-work Sat, 08 Aug 2026 04:00:00 GMT AI capacity claims matter only when a governed software factory can turn generated output into accepted, operating software. https://www.gunnertech.com/blog/shared-context-is-not-coordination https://www.gunnertech.com/blog/shared-context-is-not-coordination Sat, 08 Aug 2026 04:00:00 GMT Multi-agent software delivery needs explicit ownership, contracts, and proof because agents talking to each other does not make their combined work coherent. https://www.gunnertech.com/blog/the-agent-inbox-is-an-authority-boundary https://www.gunnertech.com/blog/the-agent-inbox-is-an-authority-boundary Sat, 08 Aug 2026 04:00:00 GMT Cross-session agent messaging needs enforceable identity, authority, and evidence gates before one worker can redirect another. https://www.gunnertech.com/blog/the-capability-gate-must-change-the-system https://www.gunnertech.com/blog/the-capability-gate-must-change-the-system Sat, 08 Aug 2026 04:00:00 GMT A dangerous model capability becomes governable only when evaluation results mechanically change access, authority, monitoring, and release conditions. https://www.gunnertech.com/blog/the-reorg-is-not-the-ai-strategy https://www.gunnertech.com/blog/the-reorg-is-not-the-ai-strategy Sat, 08 Aug 2026 04:00:00 GMT AI organizations win when research becomes governed production machinery, not when leaders redraw the reporting lines around a model. https://www.gunnertech.com/blog/the-token-bill-is-a-value-gate https://www.gunnertech.com/blog/the-token-bill-is-a-value-gate Sat, 08 Aug 2026 04:00:00 GMT Agent-run software factories should attach operating cost to business outcomes so every feature, fix, and optimization earns its route through production. https://www.gunnertech.com/blog/an-agent-credential-is-a-production-boundary https://www.gunnertech.com/blog/an-agent-credential-is-a-production-boundary Fri, 07 Aug 2026 04:00:00 GMT Agent factories need task-bound identity, short-lived credentials, and mechanical authorization that cannot expand with the conversation. https://www.gunnertech.com/blog/cheap-code-makes-selection-more-important https://www.gunnertech.com/blog/cheap-code-makes-selection-more-important Fri, 07 Aug 2026 04:00:00 GMT When agents drive implementation cost toward zero, software factories must reject changes whose proof, operation, and future drag cost more than the value they create. https://www.gunnertech.com/blog/the-factory-must-prove-who-wrote-the-code https://www.gunnertech.com/blog/the-factory-must-prove-who-wrote-the-code Fri, 07 Aug 2026 04:00:00 GMT AI-generated code needs mechanical provenance gates that can prove authorship, rights, and review before a change enters a governed codebase. https://www.gunnertech.com/blog/the-fastest-agent-is-not-the-best-factory https://www.gunnertech.com/blog/the-fastest-agent-is-not-the-best-factory Fri, 07 Aug 2026 04:00:00 GMT Agent speed and run cost matter only when the factory measures accepted outcomes, recovery, and the full workload around each attempt. https://www.gunnertech.com/blog/the-pdf-is-not-the-context https://www.gunnertech.com/blog/the-pdf-is-not-the-context Fri, 07 Aug 2026 04:00:00 GMT Agent factories should turn documents into governed, reusable context before models repeatedly pay to decode presentation files. https://www.gunnertech.com/blog/the-price-card-is-not-the-bill https://www.gunnertech.com/blog/the-price-card-is-not-the-bill Fri, 07 Aug 2026 04:00:00 GMT Model pricing becomes useful only when a software factory measures the complete cost of accepted work across routing, context, retries, verification, and recovery. https://www.gunnertech.com/blog/the-winner-expired-before-the-benchmark https://www.gunnertech.com/blog/the-winner-expired-before-the-benchmark Fri, 07 Aug 2026 04:00:00 GMT A multi-agent benchmark cannot choose your production architecture when models change faster than the operating system around them. https://www.gunnertech.com/blog/the-agent-needs-an-operating-system https://www.gunnertech.com/blog/the-agent-needs-an-operating-system Thu, 06 Aug 2026 04:00:00 GMT Advanced agentic harnesses turn model calls into governed production systems through explicit state, specialized roles, bounded concurrency, and independent proof. https://www.gunnertech.com/blog/the-cheap-model-changes-the-factory https://www.gunnertech.com/blog/the-cheap-model-changes-the-factory Thu, 06 Aug 2026 04:00:00 GMT Fast, inexpensive coding models create leverage only when software factories route work by consequence and make proof independent of model price. https://www.gunnertech.com/blog/the-job-title-hides-the-displacement https://www.gunnertech.com/blog/the-job-title-hides-the-displacement Thu, 06 Aug 2026 04:00:00 GMT AI changes software work task by task before job counts move, so leaders must redesign delivery around factories instead of waiting for a labor-market verdict. https://www.gunnertech.com/blog/a-benchmark-is-not-a-factory https://www.gunnertech.com/blog/a-benchmark-is-not-a-factory Tue, 04 Aug 2026 04:00:00 GMT Long-horizon coding benchmarks matter, but they measure a bounded reconstruction task—not whether an autonomous software factory can own change in production. https://www.gunnertech.com/blog/a-rollout-is-not-a-factory https://www.gunnertech.com/blog/a-rollout-is-not-a-factory Tue, 04 Aug 2026 04:00:00 GMT Coding-agent adoption becomes valuable only when repeatable workflows, independent proof, and durable controls turn more output into accepted change. https://www.gunnertech.com/blog/self-improvement-belongs-in-the-harness https://www.gunnertech.com/blog/self-improvement-belongs-in-the-harness Tue, 04 Aug 2026 04:00:00 GMT Agent systems improve when failures become durable tools, tests, and controls—not when a model is simply asked to reflect harder. https://www.gunnertech.com/blog/the-budget-is-part-of-the-harness https://www.gunnertech.com/blog/the-budget-is-part-of-the-harness Tue, 04 Aug 2026 04:00:00 GMT Agent budgets should govern attempts, evidence, and escalation—not merely cap a model bill after the work has already gone wrong. https://www.gunnertech.com/blog/the-evidence-can-talk-back https://www.gunnertech.com/blog/the-evidence-can-talk-back Tue, 04 Aug 2026 04:00:00 GMT AI security systems fail when hostile evidence can steer the investigator, so factories must separate observation, authority, and proof. https://www.gunnertech.com/blog/the-harness-must-not-grade-itself https://www.gunnertech.com/blog/the-harness-must-not-grade-itself Tue, 04 Aug 2026 04:00:00 GMT Self-improving agent harnesses need fixed authority, independent evidence, and a narrow editable surface before their gains mean anything. https://www.gunnertech.com/blog/the-interface-is-not-the-factory https://www.gunnertech.com/blog/the-interface-is-not-the-factory Tue, 04 Aug 2026 04:00:00 GMT A coding-agent interface earns adoption only when it improves the controls, evidence, and retained process around the work. https://www.gunnertech.com/blog/the-skill-issue-is-the-system https://www.gunnertech.com/blog/the-skill-issue-is-the-system Tue, 04 Aug 2026 04:00:00 GMT When a coding agent overbuilds, invents APIs, or chases the wrong fix, the missing capability is usually in the factory around it. https://www.gunnertech.com/blog/your-task-runner-is-an-agent-interface https://www.gunnertech.com/blog/your-task-runner-is-an-agent-interface Tue, 04 Aug 2026 04:00:00 GMT Stable repository tasks give coding agents a compact, enforceable interface instead of making every run rediscover local commands and policy. https://www.gunnertech.com/blog/keep-the-mechanism-lose-the-vampire https://www.gunnertech.com/blog/keep-the-mechanism-lose-the-vampire Mon, 03 Aug 2026 04:00:00 GMT Steve Yegge is right about agentic coding's machinery and wrong to turn one intense operating model into a universal prophecy. https://www.gunnertech.com/blog/rewriting-the-output-proves-nothing https://www.gunnertech.com/blog/rewriting-the-output-proves-nothing Mon, 03 Aug 2026 04:00:00 GMT Manual paraphrase is a weak certificate; agent-run work needs intent, evidence, provenance, and a receiver-ready decision instead. https://www.gunnertech.com/blog/your-merge-queue-is-not-a-release-strategy https://www.gunnertech.com/blog/your-merge-queue-is-not-a-release-strategy Mon, 03 Aug 2026 04:00:00 GMT Agent-scale code production needs a release system that controls risk after the merge instead of forcing every change through a human-paced gate. https://www.gunnertech.com/blog/a-label-is-not-a-requirement https://www.gunnertech.com/blog/a-label-is-not-a-requirement Sat, 01 Aug 2026 04:00:00 GMT Software factories need decisions, boundaries, and proof—not familiar labels that force agents to invent the missing product. https://www.gunnertech.com/blog/the-handoff-is-the-system https://www.gunnertech.com/blog/the-handoff-is-the-system Sat, 01 Aug 2026 04:00:00 GMT Agent fleets fail between stages unless every handoff carries a contract, evidence, and a mechanical right to refuse bad work. https://www.gunnertech.com/blog/the-factory-starves-at-the-spec https://www.gunnertech.com/blog/the-factory-starves-at-the-spec Fri, 31 Jul 2026 04:00:00 GMT When agents make implementation abundant, precise decisions become the inventory that keeps a software factory moving.