[{"data":1,"prerenderedAt":73},["ShallowReactive",2],{"blog-archive-2026-06":3},[4,24,37,51,61],{"id":5,"slug":6,"body":7,"html":8,"title":9,"description":10,"category":11,"tags":12,"author":17,"date":18,"year":19,"month":20,"quarter":21,"status":22,"featured":23},"2026\u002F06\u002Fai-in-production\u002Fevaluation-and-guardrails-before-production","evaluation-and-guardrails-before-production","\nTraditional software has a comforting property: the same input produces the same output, so a passing test suite means something. AI features don't behave that way. The same prompt can produce different answers, a model upgrade can change behaviour silently, and a content change can make a previously correct answer wrong.\n\nSo AI features need their own form of testing, **evaluation**, and it has to be a delivery gate, not a one-off exercise before a demo.\n\n## Four layers of evaluation\n\n**1. Task quality.** Does the feature do its job? For extraction, field-level accuracy against labelled documents. For classification, precision and recall per class. For summarization, coverage of required facts. For retrieval, whether the right sources come back.\n\n**2. Groundedness.** For anything generated from sources, is every claim supported by the retrieved material, and are citations correct? An ungrounded answer is a defect even when it happens to be true.\n\n**3. Safety and policy.** Does the feature refuse what it should: out-of-scope questions, requests for data the user can't access, instructions hidden in documents (prompt injection)? Does it avoid prohibited content and claims?\n\n**4. Regression.** Every change to prompts, models, retrieval settings or content is re-evaluated against the same test sets, and the results are compared with the last accepted baseline.\n\n## Building the test sets\n\nGood test sets come from the workflow, not from the vendor:\n\n- real questions and documents from the pilot, anonymized where needed\n- edge cases the business owner worries about\n- known-hard cases collected from production feedback\n- adversarial cases: injection attempts, ambiguous requests, missing data\n\nEvery case has an expected outcome defined by a person who owns the domain.\n\n## Guardrails in the application\n\nEvaluation tells you how the feature behaves. Guardrails constrain it in production:\n\n- **Grounding rules:** answer only from retrieved, authorized sources, or say you don't know.\n- **Output validation:** structured outputs checked against schemas and business rules before use.\n- **Allow-lists:** an AI can only reference entities that exist. It can't invent a product, a customer or a case number.\n- **Human checkpoints:** consequential outputs are drafts until a person accepts them.\n- **Untrusted-input handling:** document and user content is treated as data, never as instructions.\n- **Fallbacks:** if the model is unavailable or uncertain, the workflow continues deterministically.\n\n## Monitoring after launch\n\nIn production, keep measuring: acceptance and edit rates on AI drafts, user flags, drift in evaluation scores on a sampled stream, and latency and cost. Those signals feed the next round of test cases.\n\n## How the factory handles it\n\nIn our architecture, evaluation sits alongside the automated test suite. Every application foundation that includes AI features ships with an evaluation harness, and a Production Sprint doesn't close until the agreed evaluation thresholds are met. It's one of the [quality gates](\u002Fservices\u002Fai-production-sprint) we use to decide whether something is done.\n\nRelated: [from AI pilot to production application](\u002Fblog\u002Ffrom-ai-pilot-to-production-application) and [AI model governance as an application](\u002Fblog\u002Fai-model-governance-as-an-application).\n\nHave a pilot that's never been evaluated properly? [Bring it to us](\u002Fcontact).\n","\u003Cp>Traditional software has a comforting property: the same input produces the same output, so a passing test suite means something. AI features don't behave that way. The same prompt can produce different answers, a model upgrade can change behaviour silently, and a content change can make a previously correct answer wrong.\u003C\u002Fp>\n\u003Cp>So AI features need their own form of testing, \u003Cstrong>evaluation\u003C\u002Fstrong>, and it has to be a delivery gate, not a one-off exercise before a demo.\u003C\u002Fp>\n\u003Ch2>Four layers of evaluation\u003C\u002Fh2>\n\u003Cp>\u003Cstrong>1. Task quality.\u003C\u002Fstrong> Does the feature do its job? For extraction, field-level accuracy against labelled documents. For classification, precision and recall per class. For summarization, coverage of required facts. For retrieval, whether the right sources come back.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>2. Groundedness.\u003C\u002Fstrong> For anything generated from sources, is every claim supported by the retrieved material, and are citations correct? An ungrounded answer is a defect even when it happens to be true.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>3. Safety and policy.\u003C\u002Fstrong> Does the feature refuse what it should: out-of-scope questions, requests for data the user can't access, instructions hidden in documents (prompt injection)? Does it avoid prohibited content and claims?\u003C\u002Fp>\n\u003Cp>\u003Cstrong>4. Regression.\u003C\u002Fstrong> Every change to prompts, models, retrieval settings or content is re-evaluated against the same test sets, and the results are compared with the last accepted baseline.\u003C\u002Fp>\n\u003Ch2>Building the test sets\u003C\u002Fh2>\n\u003Cp>Good test sets come from the workflow, not from the vendor:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>real questions and documents from the pilot, anonymized where needed\u003C\u002Fli>\n\u003Cli>edge cases the business owner worries about\u003C\u002Fli>\n\u003Cli>known-hard cases collected from production feedback\u003C\u002Fli>\n\u003Cli>adversarial cases: injection attempts, ambiguous requests, missing data\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Every case has an expected outcome defined by a person who owns the domain.\u003C\u002Fp>\n\u003Ch2>Guardrails in the application\u003C\u002Fh2>\n\u003Cp>Evaluation tells you how the feature behaves. Guardrails constrain it in production:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>Grounding rules:\u003C\u002Fstrong> answer only from retrieved, authorized sources, or say you don't know.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Output validation:\u003C\u002Fstrong> structured outputs checked against schemas and business rules before use.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Allow-lists:\u003C\u002Fstrong> an AI can only reference entities that exist. It can't invent a product, a customer or a case number.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Human checkpoints:\u003C\u002Fstrong> consequential outputs are drafts until a person accepts them.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Untrusted-input handling:\u003C\u002Fstrong> document and user content is treated as data, never as instructions.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Fallbacks:\u003C\u002Fstrong> if the model is unavailable or uncertain, the workflow continues deterministically.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2>Monitoring after launch\u003C\u002Fh2>\n\u003Cp>In production, keep measuring: acceptance and edit rates on AI drafts, user flags, drift in evaluation scores on a sampled stream, and latency and cost. Those signals feed the next round of test cases.\u003C\u002Fp>\n\u003Ch2>How the factory handles it\u003C\u002Fh2>\n\u003Cp>In our architecture, evaluation sits alongside the automated test suite. Every application foundation that includes AI features ships with an evaluation harness, and a Production Sprint doesn't close until the agreed evaluation thresholds are met. It's one of the \u003Ca href=\"\u002Fservices\u002Fai-production-sprint\">quality gates\u003C\u002Fa> we use to decide whether something is done.\u003C\u002Fp>\n\u003Cp>Related: \u003Ca href=\"\u002Fblog\u002Ffrom-ai-pilot-to-production-application\">from AI pilot to production application\u003C\u002Fa> and \u003Ca href=\"\u002Fblog\u002Fai-model-governance-as-an-application\">AI model governance as an application\u003C\u002Fa>.\u003C\u002Fp>\n\u003Cp>Have a pilot that's never been evaluated properly? \u003Ca href=\"\u002Fcontact\">Bring it to us\u003C\u002Fa>.\u003C\u002Fp>\n","Evaluation and guardrails: how to test AI features before production","AI features need evaluation as a delivery gate, just like tests: test sets, groundedness checks, safety checks and regression on every change.","ai-in-production",[13,14,15,16],"evaluation","production","ai-governance","human-in-the-loop","fazezero-editorial","2026-06-30T00:00:00.000Z",2026,6,2,"published",false,{"id":25,"slug":26,"body":27,"html":28,"title":29,"description":30,"category":11,"tags":31,"author":17,"date":34,"year":19,"month":20,"quarter":21,"status":22,"featured":23,"series":35,"seriesOrder":36},"2026\u002F06\u002Fai-in-production\u002Ffrom-ai-pilot-to-production-application","from-ai-pilot-to-production-application","\nThere is a question we ask early in almost every conversation:\n\n> **How many AI use cases have you identified or prototyped that are not yet operating as production applications?**\n\nThe answer is rarely zero. Often it's a backlog: copilots that impressed a steering committee, agents that worked on sample data, proofs of concept that never got a production owner. The models were fine. Everything *around* the models was missing.\n\n## What a pilot proves, and what it doesn't\n\nA pilot proves that a model can do a task on representative data in a controlled setting. That is worth knowing. It does **not** prove that:\n\n- real users can reach it through enterprise identity with the right permissions\n- it integrates with the systems of record the workflow depends on\n- someone owns the output and the decisions made with it\n- failures, exceptions and edge cases have somewhere to go\n- quality is measured continuously, not once\n- the organization can audit what happened last Tuesday\n- it can be deployed, monitored and supported in the customer's cloud\n\nThose are the things production is made of.\n\n## Seven changes between pilot and production\n\n**1. From a notebook to an application.** Production AI lives inside a workflow application with a domain model, APIs and a user interface, not in a script or a chat window.\n\n**2. From shared keys to enterprise identity.** Users authenticate through the organization's identity provider. Authorization decides who can see which data and approve which actions. Multi-tenant deployments isolate data by design.\n\n**3. From sample data to integrations.** The application reads and writes through adapters to core systems, document stores and data platforms, with error handling, retries and idempotency.\n\n**4. From a demo to a human-accountable workflow.** AI drafts, summarizes, classifies and recommends. People approve, decide and remain accountable, with clear checkpoints in the workflow.\n\n**5. From one-off testing to evaluation as a gate.** Automated tests cover the application. AI evaluations cover the model behaviour, and both run on every change. See [evaluation and guardrails](\u002Fblog\u002Fevaluation-and-guardrails-before-production).\n\n**6. From “it worked” to evidence.** Every significant action produces an audit event. Controls produce evidence as a by-product of the work.\n\n**7. From a laptop to a deployment baseline.** Compute, API gateway, data, secrets, identity, AI services, observability and CI\u002FCD are defined for the target cloud.\n\n## Why most pilots stall at step two\n\nPilots are usually built to answer a capability question quickly, which is the right way to run a pilot. The trouble starts when the pilot code becomes the starting point for production. Identity, integration and controls then have to be retrofitted into a structure that never expected them, and that is where timelines collapse.\n\nWe take the opposite route. The pilot's *learning* carries forward: the prompts, the evaluation data, the workflow insight. The pilot's *code* usually doesn't. The production application starts from a deployment-ready foundation that already has steps 1, 2, 5, 6 and 7 built in, so the engineering effort goes into integration and the customer-specific workflow.\n\n## A practical path\n\n1. Pick the pilot with a named business owner and a workflow that runs weekly.\n2. Map it to the closest application foundation in the [Atlas](\u002Fatlas).\n3. Define the delta in a [Solution Definition Sprint](\u002Fservices\u002Fsolution-definition-sprint): integrations, identity, controls, evaluations and data.\n4. Build it in an [AI Production Sprint](\u002Fservices\u002Fai-production-sprint).\n5. Once it is running, add the adjacent workflows through an [Application Family Program](\u002Fservices\u002Fapplication-family-program).\n\nIf you have a backlog of pilots, [bring us the one that matters most](\u002Fcontact).\n","\u003Cp>There is a question we ask early in almost every conversation:\u003C\u002Fp>\n\u003Cblockquote>\n\u003Cp>\u003Cstrong>How many AI use cases have you identified or prototyped that are not yet operating as production applications?\u003C\u002Fstrong>\u003C\u002Fp>\n\u003C\u002Fblockquote>\n\u003Cp>The answer is rarely zero. Often it's a backlog: copilots that impressed a steering committee, agents that worked on sample data, proofs of concept that never got a production owner. The models were fine. Everything \u003Cem>around\u003C\u002Fem> the models was missing.\u003C\u002Fp>\n\u003Ch2>What a pilot proves, and what it doesn't\u003C\u002Fh2>\n\u003Cp>A pilot proves that a model can do a task on representative data in a controlled setting. That is worth knowing. It does \u003Cstrong>not\u003C\u002Fstrong> prove that:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>real users can reach it through enterprise identity with the right permissions\u003C\u002Fli>\n\u003Cli>it integrates with the systems of record the workflow depends on\u003C\u002Fli>\n\u003Cli>someone owns the output and the decisions made with it\u003C\u002Fli>\n\u003Cli>failures, exceptions and edge cases have somewhere to go\u003C\u002Fli>\n\u003Cli>quality is measured continuously, not once\u003C\u002Fli>\n\u003Cli>the organization can audit what happened last Tuesday\u003C\u002Fli>\n\u003Cli>it can be deployed, monitored and supported in the customer's cloud\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Those are the things production is made of.\u003C\u002Fp>\n\u003Ch2>Seven changes between pilot and production\u003C\u002Fh2>\n\u003Cp>\u003Cstrong>1. From a notebook to an application.\u003C\u002Fstrong> Production AI lives inside a workflow application with a domain model, APIs and a user interface, not in a script or a chat window.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>2. From shared keys to enterprise identity.\u003C\u002Fstrong> Users authenticate through the organization's identity provider. Authorization decides who can see which data and approve which actions. Multi-tenant deployments isolate data by design.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>3. From sample data to integrations.\u003C\u002Fstrong> The application reads and writes through adapters to core systems, document stores and data platforms, with error handling, retries and idempotency.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>4. From a demo to a human-accountable workflow.\u003C\u002Fstrong> AI drafts, summarizes, classifies and recommends. People approve, decide and remain accountable, with clear checkpoints in the workflow.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>5. From one-off testing to evaluation as a gate.\u003C\u002Fstrong> Automated tests cover the application. AI evaluations cover the model behaviour, and both run on every change. See \u003Ca href=\"\u002Fblog\u002Fevaluation-and-guardrails-before-production\">evaluation and guardrails\u003C\u002Fa>.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>6. From “it worked” to evidence.\u003C\u002Fstrong> Every significant action produces an audit event. Controls produce evidence as a by-product of the work.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>7. From a laptop to a deployment baseline.\u003C\u002Fstrong> Compute, API gateway, data, secrets, identity, AI services, observability and CI\u002FCD are defined for the target cloud.\u003C\u002Fp>\n\u003Ch2>Why most pilots stall at step two\u003C\u002Fh2>\n\u003Cp>Pilots are usually built to answer a capability question quickly, which is the right way to run a pilot. The trouble starts when the pilot code becomes the starting point for production. Identity, integration and controls then have to be retrofitted into a structure that never expected them, and that is where timelines collapse.\u003C\u002Fp>\n\u003Cp>We take the opposite route. The pilot's \u003Cem>learning\u003C\u002Fem> carries forward: the prompts, the evaluation data, the workflow insight. The pilot's \u003Cem>code\u003C\u002Fem> usually doesn't. The production application starts from a deployment-ready foundation that already has steps 1, 2, 5, 6 and 7 built in, so the engineering effort goes into integration and the customer-specific workflow.\u003C\u002Fp>\n\u003Ch2>A practical path\u003C\u002Fh2>\n\u003Col>\n\u003Cli>Pick the pilot with a named business owner and a workflow that runs weekly.\u003C\u002Fli>\n\u003Cli>Map it to the closest application foundation in the \u003Ca href=\"\u002Fatlas\">Atlas\u003C\u002Fa>.\u003C\u002Fli>\n\u003Cli>Define the delta in a \u003Ca href=\"\u002Fservices\u002Fsolution-definition-sprint\">Solution Definition Sprint\u003C\u002Fa>: integrations, identity, controls, evaluations and data.\u003C\u002Fli>\n\u003Cli>Build it in an \u003Ca href=\"\u002Fservices\u002Fai-production-sprint\">AI Production Sprint\u003C\u002Fa>.\u003C\u002Fli>\n\u003Cli>Once it is running, add the adjacent workflows through an \u003Ca href=\"\u002Fservices\u002Fapplication-family-program\">Application Family Program\u003C\u002Fa>.\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>If you have a backlog of pilots, \u003Ca href=\"\u002Fcontact\">bring us the one that matters most\u003C\u002Fa>.\u003C\u002Fp>\n","From AI pilot to production application: what actually changes","Most enterprise AI pilots never reach production. The gap is not the model. It is identity, integration, controls, evaluation and ownership.",[14,32,33,13],"implementation","enterprise","2026-06-25T00:00:00.000Z","the-application-factory",4,{"id":38,"slug":39,"body":40,"html":41,"title":42,"description":43,"category":44,"tags":45,"author":17,"date":49,"year":19,"month":20,"quarter":21,"status":22,"featured":23,"series":35,"seriesOrder":50},"2026\u002F06\u002Ffactory-and-architecture\u002Fcoding-assistants-are-not-an-architecture","coding-assistants-are-not-an-architecture","\nIt is the most common objection we hear from enterprises and systems integrators, and it's a fair one:\n\n> “Our developers already use Cursor, Claude Code and Copilot. Why do we need you?”\n\nThe honest answer is **not** that we have better AI. We use the same class of tools. The difference is what those tools are pointed at.\n\n## Implementation engines\n\nAI coding assistants are extraordinary at implementation. Given a clear intent and a clear structure, they write correct code quickly, refactor fluently and generate tests on request. They make good engineers much faster.\n\nWhat they don't provide is the structure itself. A general-purpose assistant has no opinion about your bounded contexts, your API versioning policy, your tenancy model or how your adapters should isolate a core banking system. It will happily produce *an* answer every time. It will not reliably produce the *same* answer every time.\n\n## Where the risk accumulates\n\nThe risks of architecture-free AI development are predictable:\n\n- **Inconsistent domain boundaries.** The same business concept lives in three services, modelled three ways.\n- **Weak API design.** Endpoints shaped by screens rather than contracts, and no versioning discipline.\n- **Late identity and authorization.** Role checks scattered through handlers, added after the demo.\n- **Inconsistent multi-tenancy.** Tenant isolation handled by convention, differently in each module.\n- **Thin testing.** Tests written to pass rather than to specify.\n- **Drift.** Regenerating a module quietly changes behaviour elsewhere.\n- **Production rework.** The prototype gets rewritten the moment it meets security review, integration or scale.\n\nNone of this is the tool's fault. It's what happens when the architecture is left to emerge.\n\n## An architectural system\n\nfazeZERO provides the part the assistants lack: a reusable application architecture refined over roughly two years and exercised across hundreds of application foundations. Concretely:\n\n- **Domain-Driven Design** with bounded contexts as the unit of structure\n- **OpenAPI-first contracts** validated before implementation\n- **Identity, authorization and multi-tenancy** generated into the core\n- **Adapters** that keep integrations out of the domain\n- **Testing and AI evaluation** as a delivery gate\n- **Observability, audit and evidence** patterns shared by every application\n- **Controlled regeneration**, so generated code can be refreshed without losing customer logic\n\nThe assistants then do what they are good at, inside those rails. We call it simply: **AI accelerates the implementation. fazeZERO governs the architecture.**\n\n## What this means for a systems integrator\n\nFor an SI, the question is not “fazeZERO or our developers.” It's whether every engagement should start with a blank repository and a fresh architecture debate. A reusable application layer means:\n\n- proposals start from an existing foundation and a defined delta\n- less pre-sales engineering on every opportunity\n- the SI's developers use their AI tools on customer-specific work, not on reinventing the plumbing\n\nWe describe the partner model in [how systems integrators can industrialize AI delivery](\u002Fblog\u002Fhow-systems-integrators-industrialize-ai-delivery).\n\n## A simple test\n\nAsk your team to build the same small application twice, a week apart, with the same assistant and the same requirements. Then compare the two architectures. If they differ in ways that would matter in production, you have found the gap an architectural system fills.\n\nSee the rails in detail on the [Architecture](\u002Ffactory\u002Farchitecture) page, or [bring us a use case](\u002Fcontact) and we'll show you the delta against the closest foundation.\n","\u003Cp>It is the most common objection we hear from enterprises and systems integrators, and it's a fair one:\u003C\u002Fp>\n\u003Cblockquote>\n\u003Cp>“Our developers already use Cursor, Claude Code and Copilot. Why do we need you?”\u003C\u002Fp>\n\u003C\u002Fblockquote>\n\u003Cp>The honest answer is \u003Cstrong>not\u003C\u002Fstrong> that we have better AI. We use the same class of tools. The difference is what those tools are pointed at.\u003C\u002Fp>\n\u003Ch2>Implementation engines\u003C\u002Fh2>\n\u003Cp>AI coding assistants are extraordinary at implementation. Given a clear intent and a clear structure, they write correct code quickly, refactor fluently and generate tests on request. They make good engineers much faster.\u003C\u002Fp>\n\u003Cp>What they don't provide is the structure itself. A general-purpose assistant has no opinion about your bounded contexts, your API versioning policy, your tenancy model or how your adapters should isolate a core banking system. It will happily produce \u003Cem>an\u003C\u002Fem> answer every time. It will not reliably produce the \u003Cem>same\u003C\u002Fem> answer every time.\u003C\u002Fp>\n\u003Ch2>Where the risk accumulates\u003C\u002Fh2>\n\u003Cp>The risks of architecture-free AI development are predictable:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>Inconsistent domain boundaries.\u003C\u002Fstrong> The same business concept lives in three services, modelled three ways.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Weak API design.\u003C\u002Fstrong> Endpoints shaped by screens rather than contracts, and no versioning discipline.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Late identity and authorization.\u003C\u002Fstrong> Role checks scattered through handlers, added after the demo.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Inconsistent multi-tenancy.\u003C\u002Fstrong> Tenant isolation handled by convention, differently in each module.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Thin testing.\u003C\u002Fstrong> Tests written to pass rather than to specify.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Drift.\u003C\u002Fstrong> Regenerating a module quietly changes behaviour elsewhere.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Production rework.\u003C\u002Fstrong> The prototype gets rewritten the moment it meets security review, integration or scale.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>None of this is the tool's fault. It's what happens when the architecture is left to emerge.\u003C\u002Fp>\n\u003Ch2>An architectural system\u003C\u002Fh2>\n\u003Cp>fazeZERO provides the part the assistants lack: a reusable application architecture refined over roughly two years and exercised across hundreds of application foundations. Concretely:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>Domain-Driven Design\u003C\u002Fstrong> with bounded contexts as the unit of structure\u003C\u002Fli>\n\u003Cli>\u003Cstrong>OpenAPI-first contracts\u003C\u002Fstrong> validated before implementation\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Identity, authorization and multi-tenancy\u003C\u002Fstrong> generated into the core\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Adapters\u003C\u002Fstrong> that keep integrations out of the domain\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Testing and AI evaluation\u003C\u002Fstrong> as a delivery gate\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Observability, audit and evidence\u003C\u002Fstrong> patterns shared by every application\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Controlled regeneration\u003C\u002Fstrong>, so generated code can be refreshed without losing customer logic\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>The assistants then do what they are good at, inside those rails. We call it simply: \u003Cstrong>AI accelerates the implementation. fazeZERO governs the architecture.\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Ch2>What this means for a systems integrator\u003C\u002Fh2>\n\u003Cp>For an SI, the question is not “fazeZERO or our developers.” It's whether every engagement should start with a blank repository and a fresh architecture debate. A reusable application layer means:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>proposals start from an existing foundation and a defined delta\u003C\u002Fli>\n\u003Cli>less pre-sales engineering on every opportunity\u003C\u002Fli>\n\u003Cli>the SI's developers use their AI tools on customer-specific work, not on reinventing the plumbing\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>We describe the partner model in \u003Ca href=\"\u002Fblog\u002Fhow-systems-integrators-industrialize-ai-delivery\">how systems integrators can industrialize AI delivery\u003C\u002Fa>.\u003C\u002Fp>\n\u003Ch2>A simple test\u003C\u002Fh2>\n\u003Cp>Ask your team to build the same small application twice, a week apart, with the same assistant and the same requirements. Then compare the two architectures. If they differ in ways that would matter in production, you have found the gap an architectural system fills.\u003C\u002Fp>\n\u003Cp>See the rails in detail on the \u003Ca href=\"\u002Ffactory\u002Farchitecture\">Architecture\u003C\u002Fa> page, or \u003Ca href=\"\u002Fcontact\">bring us a use case\u003C\u002Fa> and we'll show you the delta against the closest foundation.\u003C\u002Fp>\n","Cursor, Claude Code and Copilot are implementation engines, not architectures","Our developers already have AI coding tools, so why use fazeZERO? The answer is not better AI. It is the architectural system the AI works inside.","factory-and-architecture",[46,47,48,14],"architecture","application-factory","api-first","2026-06-23T00:00:00.000Z",3,{"id":52,"slug":53,"body":54,"html":55,"title":56,"description":57,"category":44,"tags":58,"author":17,"date":59,"year":19,"month":20,"quarter":21,"status":22,"featured":60,"series":35,"seriesOrder":21},"2026\u002F06\u002Ffactory-and-architecture\u002Fai-accelerates-architecture-governs","ai-accelerates-architecture-governs","\nEvery enterprise now has access to AI that writes code. The interesting question is no longer *whether* AI writes the application. It is **what the AI is allowed to write into**.\n\nOur answer is a fixed, proven architecture. AI accelerates the implementation inside it. It does not get to reinvent the structure for each customer.\n\n## The problem with improvised architecture\n\nGive a capable coding assistant a requirements document and a blank repository and you will get a working prototype quickly. You will also get an architecture that *emerged*: shaped by the order in which prompts arrived, the examples the model happened to favour, and whatever the developer accepted at 11pm.\n\nRun that process ten times for ten applications and you get ten architectures:\n\n- domain boundaries drawn differently every time\n- APIs designed after the screens, not before them\n- identity and authorization added once someone asks about security\n- multi-tenancy handled three different ways\n- tests written after the fact, if at all\n- code that drifts every time it is regenerated\n\nNone of this shows in the demo. All of it shows in production, as rework.\n\n## What we do instead\n\nOur pipeline runs in a fixed order:\n\n1. **Requirements**: the business problem, users, workflows and user stories.\n2. **Domain model**: bounded contexts and the vocabulary of the business.\n3. **Governed architecture**: the same layering, service boundaries and adapters every time.\n4. **API contracts**: OpenAPI definitions for every boundary, before implementation.\n5. **Generated core**: services, adapters, identity, authorization and tenancy from the factory.\n6. **Customer-specific logic and AI**: the part that is genuinely unique to this customer.\n7. **Testing and controls**: automated tests, AI evaluations, audit and evidence.\n8. **Web application and deployment**: a working UI and a baseline for the customer's cloud.\n\nAI is involved at almost every step. It drafts user stories, proposes domain models, writes implementation and generates tests. But it works *inside* rails that were set before it arrived.\n\n## Why this is faster, not slower\n\nIt sounds like governance should slow things down. In practice, it does the opposite. It removes the decisions that don't need to be made again.\n\nWhen the architecture is fixed, a new application doesn't debate how to structure identity, how to version APIs, where business rules live or how to test adapters. Those answers already exist, and they have been exercised across a large inventory of applications in very different domains. The engineering effort goes where it should: into the customer's workflow, data and integrations.\n\nThis is also why we can start customer work from an **existing application foundation** rather than a blank repository. Hundreds of foundations share one architectural grammar, so the closest one is usually a meaningful head start.\n\n## What changes and what doesn't\n\nBetween customers, **these change**: domain vocabulary, workflows, integrations, data, AI capabilities, cloud and policies.\n\n**These don't**: bounded contexts as the unit of design, contract-first APIs, identity and tenancy in the core, tests as a gate, observability, and controlled regeneration.\n\nCustomer-specific functionality changes. The architectural discipline does not.\n\n## Where this leaves AI coding tools\n\nWe use them, and so should your developers. They are excellent implementation engines. The mistake is treating them as an *architecture*. We have written more about that in [coding assistants are not an architecture](\u002Fblog\u002Fcoding-assistants-are-not-an-architecture).\n\n## Where to go next\n\n- How the pipeline works end to end: [the Application Factory](\u002Ffactory)\n- The patterns every application starts with: [Architecture](\u002Ffactory\u002Farchitecture)\n- Your own use case mapped onto it: [Solution Definition Sprint](\u002Fservices\u002Fsolution-definition-sprint)\n","\u003Cp>Every enterprise now has access to AI that writes code. The interesting question is no longer \u003Cem>whether\u003C\u002Fem> AI writes the application. It is \u003Cstrong>what the AI is allowed to write into\u003C\u002Fstrong>.\u003C\u002Fp>\n\u003Cp>Our answer is a fixed, proven architecture. AI accelerates the implementation inside it. It does not get to reinvent the structure for each customer.\u003C\u002Fp>\n\u003Ch2>The problem with improvised architecture\u003C\u002Fh2>\n\u003Cp>Give a capable coding assistant a requirements document and a blank repository and you will get a working prototype quickly. You will also get an architecture that \u003Cem>emerged\u003C\u002Fem>: shaped by the order in which prompts arrived, the examples the model happened to favour, and whatever the developer accepted at 11pm.\u003C\u002Fp>\n\u003Cp>Run that process ten times for ten applications and you get ten architectures:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>domain boundaries drawn differently every time\u003C\u002Fli>\n\u003Cli>APIs designed after the screens, not before them\u003C\u002Fli>\n\u003Cli>identity and authorization added once someone asks about security\u003C\u002Fli>\n\u003Cli>multi-tenancy handled three different ways\u003C\u002Fli>\n\u003Cli>tests written after the fact, if at all\u003C\u002Fli>\n\u003Cli>code that drifts every time it is regenerated\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>None of this shows in the demo. All of it shows in production, as rework.\u003C\u002Fp>\n\u003Ch2>What we do instead\u003C\u002Fh2>\n\u003Cp>Our pipeline runs in a fixed order:\u003C\u002Fp>\n\u003Col>\n\u003Cli>\u003Cstrong>Requirements\u003C\u002Fstrong>: the business problem, users, workflows and user stories.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Domain model\u003C\u002Fstrong>: bounded contexts and the vocabulary of the business.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Governed architecture\u003C\u002Fstrong>: the same layering, service boundaries and adapters every time.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>API contracts\u003C\u002Fstrong>: OpenAPI definitions for every boundary, before implementation.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Generated core\u003C\u002Fstrong>: services, adapters, identity, authorization and tenancy from the factory.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Customer-specific logic and AI\u003C\u002Fstrong>: the part that is genuinely unique to this customer.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Testing and controls\u003C\u002Fstrong>: automated tests, AI evaluations, audit and evidence.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Web application and deployment\u003C\u002Fstrong>: a working UI and a baseline for the customer's cloud.\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>AI is involved at almost every step. It drafts user stories, proposes domain models, writes implementation and generates tests. But it works \u003Cem>inside\u003C\u002Fem> rails that were set before it arrived.\u003C\u002Fp>\n\u003Ch2>Why this is faster, not slower\u003C\u002Fh2>\n\u003Cp>It sounds like governance should slow things down. In practice, it does the opposite. It removes the decisions that don't need to be made again.\u003C\u002Fp>\n\u003Cp>When the architecture is fixed, a new application doesn't debate how to structure identity, how to version APIs, where business rules live or how to test adapters. Those answers already exist, and they have been exercised across a large inventory of applications in very different domains. The engineering effort goes where it should: into the customer's workflow, data and integrations.\u003C\u002Fp>\n\u003Cp>This is also why we can start customer work from an \u003Cstrong>existing application foundation\u003C\u002Fstrong> rather than a blank repository. Hundreds of foundations share one architectural grammar, so the closest one is usually a meaningful head start.\u003C\u002Fp>\n\u003Ch2>What changes and what doesn't\u003C\u002Fh2>\n\u003Cp>Between customers, \u003Cstrong>these change\u003C\u002Fstrong>: domain vocabulary, workflows, integrations, data, AI capabilities, cloud and policies.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>These don't\u003C\u002Fstrong>: bounded contexts as the unit of design, contract-first APIs, identity and tenancy in the core, tests as a gate, observability, and controlled regeneration.\u003C\u002Fp>\n\u003Cp>Customer-specific functionality changes. The architectural discipline does not.\u003C\u002Fp>\n\u003Ch2>Where this leaves AI coding tools\u003C\u002Fh2>\n\u003Cp>We use them, and so should your developers. They are excellent implementation engines. The mistake is treating them as an \u003Cem>architecture\u003C\u002Fem>. We have written more about that in \u003Ca href=\"\u002Fblog\u002Fcoding-assistants-are-not-an-architecture\">coding assistants are not an architecture\u003C\u002Fa>.\u003C\u002Fp>\n\u003Ch2>Where to go next\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>How the pipeline works end to end: \u003Ca href=\"\u002Ffactory\">the Application Factory\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>The patterns every application starts with: \u003Ca href=\"\u002Ffactory\u002Farchitecture\">Architecture\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>Your own use case mapped onto it: \u003Ca href=\"\u002Fservices\u002Fsolution-definition-sprint\">Solution Definition Sprint\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n","AI accelerates the implementation. Architecture governs it.","Why fazeZERO uses AI inside a proven production architecture instead of letting it improvise a new architecture for every enterprise application.",[47,46,14,33],"2026-06-18T00:00:00.000Z",true,{"id":62,"slug":63,"body":64,"html":65,"title":66,"description":67,"category":68,"tags":69,"author":17,"date":71,"year":19,"month":20,"quarter":21,"status":22,"featured":23,"series":35,"seriesOrder":72},"2026\u002F06\u002Fpartners-and-channel\u002Fhow-systems-integrators-industrialize-ai-delivery","how-systems-integrators-industrialize-ai-delivery","\nSystems integrators are being asked the same question by every client: *how do we get our AI use cases into production?* Most SIs have strong customer access, cloud and data practices, and growing AI teams. What few have is an **industrialized application layer**: a repeatable way to turn a use case into a governed application without starting from a blank repository each time.\n\nThat is the gap a joint AI application factory fills.\n\n## The operating thesis\n\n> **The SI discovers, sells, integrates and operates. fazeZERO productizes and builds.**\n\nThe split follows what each side does best.\n\n**The systems integrator owns:**\n\n- the customer relationship and business context\n- enterprise discovery\n- cloud, infrastructure and data\n- enterprise integrations and cybersecurity\n- managed services, change and operations\n\n**fazeZERO owns:**\n\n- product definition, architecture and domain modeling\n- the application factory and code generation\n- application-level AI\n- testing and reusable engineering\n- application-specific architecture\n\nNobody competes for the same work, and the customer gets both an application and someone to run it.\n\n## Why it raises conversion\n\nThe hardest moment in an AI opportunity is the gap between an enthusiastic workshop and a funded project. The customer asks what exactly they would get, how long it would take and how risky it is. Without a reusable application layer, answering means weeks of unpaid pre-sales engineering.\n\nWith one, the answer starts from an existing application foundation and a defined delta:\n\n1. **Opportunity qualification.** Account context, application matching and discovery questions. Included.\n2. **Joint discovery workshop.** Use-case selection and an architecture hypothesis, within an agreed pre-sales allowance.\n3. **AI Opportunity Brief.** A joint deliverable, so the SI leaves the room with something concrete to sell.\n4. **Solution Definition Sprint.** Formal requirements, domain model, integrations and scope. Billable.\n\nThe rule we follow: **we don't charge partners to bring us into the room. We charge once the conversation turns into solution engineering.**\n\n## Rules that protect the account\n\nPartnerships fail on ambiguity, so the rules of engagement are explicit:\n\n- **The originator keeps the relationship.** On partner-originated accounts, the partner is normally the commercial prime.\n- **Registration and protection.** Opportunities are registered (account, use case, sponsor, owner, stage) and protected while active.\n- **No circumvention.** Neither party pursues a registered opportunity around the other.\n- **No blanket exclusivity.** Preferred status is earned through bookings, delivery and commitment.\n- **Clear IP.** The factory, generator and reusable assets stay with fazeZERO. Customer-specific IP follows the customer contract.\n- **Co-branding over white label.** “Your AI Application Factory, powered by fazeZERO” builds credibility for both sides.\n\n## The tools\n\nPartners work in a private workbench on top of the [Application Atlas](\u002Fatlas): account mapping, problem-to-application matching, discovery questions, opportunity briefs and opportunity registration. The workbench is one configurable codebase with partner branding. We don't build a separate fork for each partner.\n\n## How to start\n\nStart small and measurable: ten named accounts, a handful of repeatable AI opportunities, three to five joint discovery sessions and one paid sprint. The goal of the first quarter isn't only revenue. It's learning whether the partner can generate qualified opportunities without us originating every one.\n\nRead more on [systems integrator partnerships](\u002Fpartners\u002Fsystems-integrators), or [start the conversation](\u002Fcontact).\n","\u003Cp>Systems integrators are being asked the same question by every client: \u003Cem>how do we get our AI use cases into production?\u003C\u002Fem> Most SIs have strong customer access, cloud and data practices, and growing AI teams. What few have is an \u003Cstrong>industrialized application layer\u003C\u002Fstrong>: a repeatable way to turn a use case into a governed application without starting from a blank repository each time.\u003C\u002Fp>\n\u003Cp>That is the gap a joint AI application factory fills.\u003C\u002Fp>\n\u003Ch2>The operating thesis\u003C\u002Fh2>\n\u003Cblockquote>\n\u003Cp>\u003Cstrong>The SI discovers, sells, integrates and operates. fazeZERO productizes and builds.\u003C\u002Fstrong>\u003C\u002Fp>\n\u003C\u002Fblockquote>\n\u003Cp>The split follows what each side does best.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>The systems integrator owns:\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cul>\n\u003Cli>the customer relationship and business context\u003C\u002Fli>\n\u003Cli>enterprise discovery\u003C\u002Fli>\n\u003Cli>cloud, infrastructure and data\u003C\u002Fli>\n\u003Cli>enterprise integrations and cybersecurity\u003C\u002Fli>\n\u003Cli>managed services, change and operations\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cstrong>fazeZERO owns:\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cul>\n\u003Cli>product definition, architecture and domain modeling\u003C\u002Fli>\n\u003Cli>the application factory and code generation\u003C\u002Fli>\n\u003Cli>application-level AI\u003C\u002Fli>\n\u003Cli>testing and reusable engineering\u003C\u002Fli>\n\u003Cli>application-specific architecture\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Nobody competes for the same work, and the customer gets both an application and someone to run it.\u003C\u002Fp>\n\u003Ch2>Why it raises conversion\u003C\u002Fh2>\n\u003Cp>The hardest moment in an AI opportunity is the gap between an enthusiastic workshop and a funded project. The customer asks what exactly they would get, how long it would take and how risky it is. Without a reusable application layer, answering means weeks of unpaid pre-sales engineering.\u003C\u002Fp>\n\u003Cp>With one, the answer starts from an existing application foundation and a defined delta:\u003C\u002Fp>\n\u003Col>\n\u003Cli>\u003Cstrong>Opportunity qualification.\u003C\u002Fstrong> Account context, application matching and discovery questions. Included.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Joint discovery workshop.\u003C\u002Fstrong> Use-case selection and an architecture hypothesis, within an agreed pre-sales allowance.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>AI Opportunity Brief.\u003C\u002Fstrong> A joint deliverable, so the SI leaves the room with something concrete to sell.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Solution Definition Sprint.\u003C\u002Fstrong> Formal requirements, domain model, integrations and scope. Billable.\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>The rule we follow: \u003Cstrong>we don't charge partners to bring us into the room. We charge once the conversation turns into solution engineering.\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Ch2>Rules that protect the account\u003C\u002Fh2>\n\u003Cp>Partnerships fail on ambiguity, so the rules of engagement are explicit:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>The originator keeps the relationship.\u003C\u002Fstrong> On partner-originated accounts, the partner is normally the commercial prime.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Registration and protection.\u003C\u002Fstrong> Opportunities are registered (account, use case, sponsor, owner, stage) and protected while active.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>No circumvention.\u003C\u002Fstrong> Neither party pursues a registered opportunity around the other.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>No blanket exclusivity.\u003C\u002Fstrong> Preferred status is earned through bookings, delivery and commitment.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Clear IP.\u003C\u002Fstrong> The factory, generator and reusable assets stay with fazeZERO. Customer-specific IP follows the customer contract.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Co-branding over white label.\u003C\u002Fstrong> “Your AI Application Factory, powered by fazeZERO” builds credibility for both sides.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2>The tools\u003C\u002Fh2>\n\u003Cp>Partners work in a private workbench on top of the \u003Ca href=\"\u002Fatlas\">Application Atlas\u003C\u002Fa>: account mapping, problem-to-application matching, discovery questions, opportunity briefs and opportunity registration. The workbench is one configurable codebase with partner branding. We don't build a separate fork for each partner.\u003C\u002Fp>\n\u003Ch2>How to start\u003C\u002Fh2>\n\u003Cp>Start small and measurable: ten named accounts, a handful of repeatable AI opportunities, three to five joint discovery sessions and one paid sprint. The goal of the first quarter isn't only revenue. It's learning whether the partner can generate qualified opportunities without us originating every one.\u003C\u002Fp>\n\u003Cp>Read more on \u003Ca href=\"\u002Fpartners\u002Fsystems-integrators\">systems integrator partnerships\u003C\u002Fa>, or \u003Ca href=\"\u002Fcontact\">start the conversation\u003C\u002Fa>.\u003C\u002Fp>\n","How systems integrators can industrialize AI delivery","A joint AI application factory: the SI owns the customer, integration and operations while fazeZERO productizes and builds the application layer.","partners-and-channel",[70,47,32,33],"partners","2026-06-16T00:00:00.000Z",1,1790080513985]