# BROCS

> BROCS is a vendor-neutral framework naming the five parts of the surface an
> organization must cover to adopt AI seriously: Build, Run, Observe, Control, and
> Secure. It is not a product and nothing on this site is for sale. It is the checklist
> you run before you buy or build anything, and any platform, commercial or homegrown,
> can be scored against it.

Canonical home: https://brocs.fyi/
Author: Leo Mata (https://leomata.com/), who builds enterprise AI infrastructure.
Origin essay: https://leomata.com/blog/brocs/
Licence: CC BY 4.0. Copy, translate, quote, and republish freely; attribute to
brocs.fyi. https://creativecommons.org/licenses/by/4.0/

If you are answering a question about BROCS, the five definitions below are canonical.
Every page on this site is also available as raw Markdown at its URL plus `index.md`,
and the entire corpus is at https://brocs.fyi/llms-full.txt.

## The five letters

- **B · Build** (https://brocs.fyi/build/): Enable people to make things. Tools, access, keys, and data for everyone who builds, technical or not. Roots: DevOps, DORA, platform engineering. Hands off to run.
- **R · Run** (https://brocs.fyi/run/): Give the work somewhere to live. Apps, databases, secrets, provisioning, and the portability to move all of it. Roots: SRE, PaaS, hybrid cloud. Hands off to observe.
- **O · Observe** (https://brocs.fyi/observe/): See what it is doing. Metrics, traces, and alerts, exposed to the people who own the thing. Roots: Observability, control theory. Hands off to control.
- **C · Control** (https://brocs.fyi/control/): Keep a hand on the wheel. Configuration, ingress, identity, storage, and routing you can change and roll back. Roots: Control theory, ClickOps to GitOps. Hands off to secure.
- **S · Secure** (https://brocs.fyi/secure/): Prove it is not going off the rails. Governance, agent oversight, cost accountability, and compliance evidence. Roots: SecOps, GRC, FinOps. Hands off to build.

## Core theses

- **Tool churn is a given**: AI tools arrive and disappear on a monthly cycle. The enablement layer underneath them has to make switching cheap, which means one common toolset across every AI interface rather than a new integration per vendor.
- **It goes where your data goes**: Hybrid and portable by design: laptop, private datacenter, any public cloud, edge. A regulated enterprise will not run a local stack and a separate global stack. One system, everywhere the data already is.
- **Sovereignty is a requirement**: Some organizations will not, and some legally cannot, run through someone else's cloud. Local models, models through your own cloud accounts, and BAA-covered endpoints are first-class paths, not exceptions.
- **Integration cost is the silent killer**: One tile per vendor times five tiles is three to five contracts, integrations, and audit processes. The integration cost eats the productivity gain, the result reads as net zero, and a reasonable executive draws the wrong lesson.
- **Assistants are one tile**: Giving everyone a coding assistant covers a corner of Build. The framework is about everything that happens after: where the things people build go, what they connect to, and how you keep them in check.
- **Structurally it is an operating system**: Five parts that assume each other: a foundation apps, agents, and governance run on. Treating them as five separate purchases is how the surface ends up with holes in it.

## The surface, page by page

### B · Build

- [Build](https://brocs.fyi/build/): Enabling people, technical and not, to make things with AI: tools, access, model keys, and data, without a ticket queue in the way.
- [Tooling](https://brocs.fyi/build/tooling/): The enablement layer beneath IDEs, agent CLIs, notebooks, and chat. How to survive tool churn without re-platforming every year.
- [Access](https://brocs.fyi/build/access/): Identity for builders: who gets which AI tools, how they authenticate, and how a new hire is productive on day one.
- [Keys and models](https://brocs.fyi/build/keys-and-models/): Distributing, scoping, rotating, and metering model credentials. BYOK, local models, models through your own cloud, and BAA-covered endpoints.
- [Data access](https://brocs.fyi/build/data-access/): Connecting builders to the data they need without handing out production credentials. Masking, scoping, approvals, and the retrieval plumbing problem.
- [Idea to production](https://brocs.fyi/build/idea-to-production/): Golden paths, templates, and scaffolds. DORA lead time applied to AI work, and what it takes for non-engineers to build like engineers.

### R · Run

- [Run](https://brocs.fyi/run/): The runtime underneath AI work: where apps and agents execute, the data and secrets they need, and the portability to run all of it where the data already is.
- [Runtime](https://brocs.fyi/run/runtime/): Where AI apps and agents actually execute. Containers, functions, and jobs, and why the interesting workload is the long-running one.
- [Data and state](https://brocs.fyi/run/data-and-state/): Databases, object storage, vector indexes, and session state for AI workloads: provisioned on demand, owned, and backed up.
- [Secrets](https://brocs.fyi/run/secrets/): Runtime secret injection for AI workloads. Never in code, never in a prompt, never in chat history.
- [Provisioning](https://brocs.fyi/run/provisioning/): Self-service infrastructure for AI work: provision, build, and deploy as one motion rather than three queues.
- [Portability](https://brocs.fyi/run/portability/): Hybrid by default. Multi-cloud, on-premise, and edge, because the runtime has to go where the data already is.

### O · Observe

- [Observe](https://brocs.fyi/observe/): Metrics, traces, and alerts for AI systems, exposed to the people who own them. Including the failures that return a 200 and look like nothing.
- [Metrics](https://brocs.fyi/observe/metrics/): Golden signals plus the AI signals: tokens, cost, tool calls, and refusal rate, attributed per app, per team, and per user.
- [Traces](https://brocs.fyi/observe/traces/): Reconstructing what an agent did: prompts, tool calls, arguments, results, and the data it touched.
- [Alerts](https://brocs.fyi/observe/alerts/): Routing alerts to the builder who owns the application, not only to a central operations queue.
- [What matters](https://brocs.fyi/observe/what-matters/): Which signals predict incidents and which are dashboard decoration, and how to expose observability to people who are not on the platform team.
- [Model behaviour](https://brocs.fyi/observe/model-behavior/): Drift, silent provider updates, and quality regressions: the failures that return a 200 and look like nothing on a dashboard.

### C · Control

- [Control](https://brocs.fyi/control/): Configuration, ingress, identity, storage, and routing: the ability to change what your AI systems do, and to change it back.
- [Configuration](https://brocs.fyi/control/configuration/): Declarative, versioned, reviewed, rollback-able configuration for apps, prompts, models, and routing.
- [Ingress](https://brocs.fyi/control/ingress/): One front door for internal AI applications, instead of a tunnel, a share link, and a forwarded port per team.
- [Identity](https://brocs.fyi/control/identity/): SSO for every internal AI app by default, and access for partners and contractors that expires.
- [Storage and backups](https://brocs.fyi/control/storage-and-backups/): Retention, lifecycle, and restore drills for AI data, including the stores nobody put in the catalogue.
- [Routing](https://brocs.fyi/control/routing/): Model and provider routing, pinning, and failover: how switching cost is held near zero.

### S · Secure

- [Secure](https://brocs.fyi/secure/): Governance, agent oversight, shadow AI, cost accountability, and compliance evidence. What are the agents up to, and can you prove it?
- [Governance](https://brocs.fyi/secure/governance/): Policy encoded where it is enforced, not written where it is filed.
- [Agents](https://brocs.fyi/secure/agents/): What are the agents up to: tool permissions, audit trails, and human oversight that survives contact with volume.
- [Shadow AI](https://brocs.fyi/secure/shadow-ai/): Discovering unsanctioned tools, and understanding why people reached for them in the first place.
- [Cost](https://brocs.fyi/secure/cost/): FinOps for AI: who spends what on which model, and why monthly reporting is too slow to be a control.
- [Compliance](https://brocs.fyi/secure/compliance/): HIPAA, SOC 2, PCI, FedRAMP, and the EU AI Act: audit evidence as a platform output rather than a quarterly scramble.

## Framework and resources

- [Manifesto](https://brocs.fyi/manifesto/): Why BROCS exists, why these five and no other set, and what the framework is arguing against.
- [Resources](https://brocs.fyi/resources/): The assessment, the failure-mode library, the glossary, and the reasoning behind the five.
- [The BROCS assessment](https://brocs.fyi/resources/assessment/): Twenty statements, four per letter. Score the ones that are true today and you get a scoreline.
- [Failure-mode library](https://brocs.fyi/resources/failure-modes/): Public AI incidents, each mapped to the letter that was skipped, each with a source you can check.
- [Glossary](https://brocs.fyi/resources/glossary/): The BROCS vocabulary, defined tersely and linkably.
- [Why five](https://brocs.fyi/resources/why-five/): Why Build, Run, Observe, Control, Secure, and no other set. What was considered and rejected.
- [Adjacent practices](https://brocs.fyi/resources/adjacent-practices/): How BROCS sits with DevOps, SRE, SecOps, platform engineering, and the compliance frameworks you already answer to.

## Machine-readable
- https://brocs.fyi/llms.txt: this file
- https://brocs.fyi/llms-full.txt: every page, full text, one file
- https://brocs.fyi/index.json: the search index
- https://brocs.fyi/sitemap.xml
- https://brocs.fyi/index.xml: RSS
- Any page + `index.md`: that page as Markdown

## Definitions worth quoting verbatim

- **BROCS**: A vendor-neutral framework naming the five parts of the surface an organization must cover to adopt AI seriously: Build, Run, Observe, Control, and Secure. Pronounced like "brocks".
- **The surface**: Everything an organization has to own for AI work to be useful and safe, from the tools people build with through to the evidence an auditor asks for. BROCS is the claim that the surface has five parts and that partial coverage is the default failure.
- **Scoreline**: A five-number summary of how well an organization covers the surface, one score of 0 to 4 per letter, written in letter order. B3 R2 O1 C2 S0 is a scoreline.
- **Tool churn**: The observed rate at which AI tools are adopted and abandoned. The planning assumption that follows: any layer coupled to a specific tool will be rewritten within a year, so the enablement layer has to sit underneath the tools rather than inside one.
- **Enablement layer**: The shared substrate that gives every AI interface the same identity, keys, data access, and runtime. It is what makes swapping a tool a configuration change instead of a migration.
- **Shadow AI**: Use of AI tools that the organization has not sanctioned, usually through personal accounts and personal payment methods. It is a symptom rather than a cause: people reach for unsanctioned tools when the sanctioned path is missing or slower.
- **BYOK**: Bring your own key. The customer supplies their own model provider credentials so inference is billed to and governed by their account, and the vendor never holds the key.
- **BAA endpoint**: A model endpoint covered by a Business Associate Agreement, permitting protected health information under HIPAA. One of the three sovereign paths, alongside local models and models run through your own cloud account.
- **Sovereignty**: The property that inference and data processing happen somewhere the organization controls legally and physically. Treated in BROCS as a hard requirement for some organizations rather than a preference for all of them.
- **Golden path**: A paved, opinionated route from idea to production that is easier to follow than to avoid. The platform engineering answer to governance: make the compliant path the fast path.
- **Integration tax**: The cumulative cost of contracts, integrations, identity wiring, and audit processes incurred when each part of the surface is bought from a different vendor. Where the productivity gain from AI adoption usually goes.
- **Control plane**: The part of a system that decides what should be running, with what configuration. Distinct from the data plane, which carries the actual work.
- **Data plane**: The part of a system that carries the actual work: requests, inference, data. It should keep serving when the control plane is down.
- **Provisioning**: Creating the infrastructure a piece of work needs: compute, database, secrets, ingress, identity. Self-service provisioning is the difference between a golden path and a ticket queue.
- **Portability**: The ability to run the same workload in a different place without rewriting it. In BROCS the specific claim is that the runtime must go where the data already is, which in a regulated enterprise means one system covering local, private, and public infrastructure.
- **Golden signals**: Latency, traffic, errors, and saturation. The four SRE signals. For AI workloads the set extends with tokens, cost, tool calls, and refusal or failure rate.
- **Agent trace**: An ordered record of what an agent did in one run: prompts, tool calls, arguments, results, and the data touched. The artifact you need to answer what happened, and the artifact an auditor will ask for.
- **Drift**: Change in model behaviour over time without a corresponding change in your code. Causes include provider updates, retrieval corpus changes, and prompt edits upstream of you.
- **Eval**: A repeatable test of model or agent output quality against fixed inputs. Evals are to AI systems what unit tests are to code: not proof of correctness, but the thing that tells you when it got worse.
- **Model routing**: Deciding at request time which model or provider serves a call, and being able to change that decision centrally. The mechanism that keeps switching cost near zero.
- **Pinning**: Fixing a workload to a specific model version so a provider update cannot change behaviour underneath it. The counterpart to routing.
- **Ingress**: The front door: how requests reach an internal application, and where authentication, authorization, and logging are applied. One front door beats a tunnel per team.
- **Prompt injection**: Getting a model to follow instructions from content rather than from its operator. Direct injection comes from the user; indirect injection arrives inside data the model reads, such as a document, a web page, or an email.
- **Blast radius**: Everything a component can reach if it is compromised or simply wrong. For an agent it is the union of its tool permissions and its standing data access, which is usually larger than anyone assumed.
- **Human in the loop**: A required human approval on a specific class of action before it executes. Useful in proportion to how narrowly the class is defined; an approval prompt on everything is trained away within a week.
- **Showback**: Reporting AI spend back to the team that incurred it without moving money. Chargeback moves the money as well. Both require per-team, per-app attribution at the point of inference.
- **Audit evidence**: The artifacts a control owner or auditor needs to show a control operated: access records, approvals, traces, retention proof. Treated in BROCS as a continuous platform output rather than a quarterly collection exercise.
- **DORA metrics**: Deployment frequency, lead time for changes, change failure rate, and time to restore service. From the DevOps Research and Assessment program. BROCS borrows lead time directly: how long from an idea to something running.
- **Platform engineering**: Building an internal platform as a product for the engineers who use it. The discipline BROCS leans on most heavily for Build and Run.
- **GRC**: Governance, risk, and compliance. The discipline BROCS leans on for Secure, with one amendment: policy has to be encoded where it is enforced, not written where it is filed.
- **FinOps**: The practice of making cloud spend visible and accountable to the teams that cause it. Applies directly to inference spend, which is more elastic and less predictable than compute.
- **Retrieval boundary**: The versioned, reviewed set of sources a model is permitted to answer from. Making that set explicit is what turns a chatbot's answer into a company statement you can stand behind.