Claude Code for business: what a real deployment involves

Deploying Claude Code across a business is not an install. It is eight decisions: where it runs, who can reach it, what it is allowed to touch, what staff can override, how it is extended, how it is secured, how you measure what it costs, and who owns it in six months. Get the install right and the decisions wrong and you have a clever tool one person uses, not a capability the business has.

That framing is not ours. It is roughly how Anthropic itself assesses the people it badges to deploy Claude Code for clients, and this week it issued us that badge: the Claude Partner Badge for Claude Code, verifiable on Credly.

Below is what the assessment actually asks for, and the two artefacts from it that are worth more to you than the badge is.

What is the Claude Code partner badge?

It is a credential issued by Anthropic through Credly to partners who complete its Claude Code enablement path. Eight courses and a scenario-based capstone assessment. It is not a proctored exam, and it is worth being precise about that, because a lot of AI credentials are thinner than they sound.

What makes it useful is that it makes you hand in real work rather than tick boxes: configuration packs, a deployment decision document, a completed security questionnaire, custom commands, Skills, hooks, telemetry configuration. The subject is not "can you use Claude Code". It is "could you put Claude Code inside somebody else's business and stand behind it".

The eight modules split cleanly in half. Four are what you would expect: product foundations, installation and environments, deployment architecture, configuration and extensibility. The other four are the ones almost nobody posts about, and they are the four a business owner should actually care about.

The four modules that decide whether it survives contact with a business

Deployment architecture. Where it runs, who can reach it, what it touches. The difference between one person using a clever tool on a laptop and a business having a capability.

Security and governance. Managed settings your staff cannot override, single sign-on, user provisioning and de-provisioning, the vendor's actual compliance position, and how you answer a client's security questionnaire when they send you one. If you sell into anyone regulated, this is the module that decides whether you win the contract.

Administration and measurement. Audit trails, usage telemetry, cost per team. Not glamorous. It is the difference between a deployment you can defend at a board meeting and one you quietly stop mentioning.

Delivery methodology. How you scope it, roll it out, train people on it, and hand it over so it is still running in six months.

That last one is where most AI adoption stories end. The pilot works. The demo lands. Then nobody owns it, nobody can see what it cost, nobody wrote down who is allowed to use it for what, and eleven months later the business is back where it started with a slightly worse opinion of AI.

Artefact one: what a Claude Code deployment decision document contains

This is the document the capstone asks for, and the one we now write before any client rollout. If you are doing this yourself, this is the whole thing in outline. Every line is a decision somebody has to make, and writing down who made it is half the value.

• Where it runs. Individual laptops, a shared always-on machine, or cloud. Decides your blast radius and your backup story.
• Which accounts. Company-owned identities only, never personal accounts. This is the single most common failure in small business AI adoption.
• What it can reach. Repositories, drives, inboxes, databases, third party tools. List them explicitly. Anything not on the list is denied.
• What it can do unattended. Read, draft and log are usually fine unattended. Send, pay, publish and delete are not. Draw that line before rollout, not after an incident.
• What staff cannot override. Managed settings enforced centrally, not a policy document people are asked to remember.
• Who approves what. Named humans against named actions, with the approval recorded.
• How it is extended. Which integrations, which custom commands, which reusable skills, and who is allowed to add one.
• What is logged. Every action, who or what triggered it, and where the log lives.
• What it costs and who watches. Usage exported to a dashboard from day one, reviewed monthly against a threshold.
• Handover. Who owns it, what the runbook is, and what happens when that person leaves.

Artefact two: the security questions you will be asked

The capstone includes completing a client security questionnaire, which is a useful exercise because these are the questions a real buyer sends you. If you are evaluating any AI supplier, including us, these are fair to ask and you should expect written answers.

• Does the vendor train on our data, and on which plan tier does that change?
• Is there a data processing agreement, and what does it actually cover?
• Where is data processed and retained, and for how long?
• How do we provision and, more importantly, de-provision a leaver?
• What can an individual user change, and what is locked centrally?
• Is there an audit trail, who can read it, and can it be exported?
• What is the vendor's compliance position, and can they evidence it?
• What is the incident process, and who calls it?

If a supplier cannot answer those in writing, that is the answer.

Why we sat it

We are not learning this on your time. Assist IQ runs its own operations on Claude Code, all day and every day, and has done for months. Our will writing business runs a production pipeline on the same stack. The badge is not where the experience comes from.

What it is, is a way of showing an owner, before they have to take a leap of faith, that the person walking into their business has been assessed on the governance and delivery half rather than just the fun half. Alongside it sits the Claude Certified Associate credential, also from Anthropic, also verifiable. Both are on our about page with their verification links, because a badge you cannot click is not a credential, it is a picture.

What this means if you are the owner

Very little on its own. Credentials do not do the work.

What it should tell you is what to ask of anyone you are considering, us included. Ask where it runs. Ask what your staff can and cannot override. Ask what happens to your data and who has agreed to what in writing. Ask what it costs per person per month and how you would know. Ask what happens when the person who set it up leaves.

If they cannot answer those, the demo was the product.

Two related pieces worth reading first, both written from the vendors' own published policies and free to lift for your own staff policy: Is AI safe for my business data? and ChatGPT and Claude: free vs business vs enterprise vs API.

And if you want to see what a governed deployment looks like inside a business of five to fifty people, that is what Claude for Business is, and what an AI employee does once it is in. Happy to walk you through it with no pitch attached.

Questions we get asked

What is the Claude Partner Badge for Claude Code?

It is a credential issued by Anthropic through Credly to partners who complete its Claude Code enablement path: eight courses plus a scenario-based capstone assessment. It is not a proctored exam. The assessment requires real deliverables including configuration packs, a deployment decision document, a completed security questionnaire, custom commands, Skills, hooks and telemetry configuration, so it tests whether you can scope, deploy, configure and run Claude Code inside a client business rather than just use it yourself.

Do you need developers or an IT department to run Claude Code in a business?

No, but you do need someone to own the deployment. Claude Code is a command line tool, so the install and configuration work is technical, but the day to day use is written instructions rather than code. The parts that need ownership are access control, managed settings, approval rules for anything that touches money or customers, and cost monitoring. Most owner-led businesses bring that in rather than hire for it.

Is Claude Code safe for company data?

On a business or enterprise plan, Anthropic does not train on your data by default and a data processing agreement comes with the contract. The real risk in most deployments is not the vendor, it is configuration: staff running it on personal accounts, no audit trail, no restrictions on what it can reach, and no record of who approved what. Those are configuration decisions you control, and they are the ones worth writing down before you roll anything out.

How much does Claude Code cost to run across a team?

It depends on usage rather than seats, which is exactly why cost telemetry matters. Anthropic supports OpenTelemetry export, so usage and spend can be tracked per user and per team from day one. A deployment without that reporting will produce a bill nobody can explain, which is usually the point at which the business quietly stops using it.

What should I ask an AI consultant before hiring them?

Ask where it runs, what your staff can and cannot override, what happens to your data and who has agreed to what in writing, what it costs per person per month and how you would know, and what happens when the person who set it up leaves. If they cannot answer those five questions, the demo was the product.