ABC Legal, a US legal document delivery company with 1,100 employees, rolled out Claude Enterprise earlier this year. Staff across service of process, eFiling, compliance, finance and marketing started building their own automations without being asked. That is the adoption every business owner hopes for. It is also exactly where most Australian businesses we talk to get stuck: pilots everywhere, no one owns them, and no one can say what they cost or whether they ran last night.
ABC Legal's CTO, Brandon Fuller, solved this by moving from ad hoc scheduled tasks on individual laptops to Claude Managed Agents: one deployment structure, shared workspaces, a single audit and billing surface, and agents that run unattended in the cloud instead of on someone's desktop. As of July 2026 the numbers read: more than 50 agents in production, up to roughly a 50% reduction in the cost of the human tasks some agents cover, and around 310 employees using Claude for daily work.
The pattern worth copying: agent-as-code
Every agent at ABC Legal is defined as code. A prompt plus configuration lives in a git repository the whole company can see, review and improve. Nothing about an agent changes except through a pull request someone approves, which gives every agent version history, code review, rollback and an audit trail. Fuller built two starter templates: one for event-driven agents that fire the moment something happens, and one for scheduled agents that run on a timer. A 15-person steering committee of non-developers cloned the repo, copied a template, and told Claude Code what the agent should do. Within a week all 15 had working agents. Within a month, roughly 50 were running across the company.
Three examples give a sense of the range:
An AI code reviewer that runs multi-model analysis across four codebases before engineers merge.
An evidence-delivery agent that replaced a weekly manual export-and-upload chore, built in about an hour by an account manager who had never automated anything before.
An eFiling rejection diagnoser that reads a court's rejection, checks the court's rules, and posts a diagnosis to Slack in about a minute.
Why this matters for AU mid-market
We see the same failure mode with almost every business that adopts Claude on its own: someone in ops builds a brilliant scheduled task, it works, and then it lives on their laptop with no owner, no version history and no way to know if it silently broke. ABC Legal's fix is not exotic. It is treating an agent like software from day one: named, owned, reviewed, and running somewhere the whole business can see.
For an Australian business with 50 to 500 staff, standing up that same governed structure, a starter template, a shared agent registry, and an audit trail your finance and compliance team can actually read, is typically a two-to-four week scoped build, in the order of $8,000 to $18,000 depending on how many agent types you are consolidating and what systems they touch. That is a small number against the cost of a critical automation failing silently because nobody owned it.
The signal that it is time to move
There is a simple test for whether you have outgrown pilots. If you cannot produce a single list of every Claude automation running in your business, who owns each one, and what it costs, you are past the point where ad hoc tasks on laptops are safe. That is not a reason to slow down adoption; it is the reason to give it a spine. Move the scattered wins into one governed structure and you keep the enthusiasm while removing the risk that made it fragile.
How to start without boiling the ocean
You do not need to consolidate fifty agents on day one. Start with a registry: a single shared list of every Claude automation running, who owns it, and what it does. That alone surfaces the orphaned tasks and the silent failures. Then take the two or three that matter most, define them as code in a shared repository with a template, and move them off their laptops into a place the whole business can see. Prove the pattern on a handful, then widen it.
Done this way the change is incremental and low-risk. You are not pausing the work your team already values; you are giving your best automations a name, an owner and a version history so they stop being fragile. The businesses that scale Claude well are rarely the ones with the cleverest single agent. They are the ones that treated agents like software early, before the sprawl became the problem.
If you have Claude automations scattered across your team and no one can say what is running or what it costs, book a session and we will scope the move from pilot to managed.
