Agentic Digital Identity details controls for AI agent
Wed, 18th Nov 2026 (Today)
Agentic Digital Identity has outlined the controls it uses to govern its own AI agent after more than five months of running day-to-day operations on the system. The same controls are included in the agents it builds for clients.
The Australian firm said its internal agent, tronic, drafts outbound email, turns meeting transcripts into minutes and action items, runs approved tasks from a project board and reports into Slack. Founder Thomas Beard said the main effort since launch has focused not on the language model itself, but on the mechanisms around it that determine what the system can do and how its actions are verified.
"Plain code does the work and the model handles the language," said Beard, Engineer, Agentic Digital Identity. "A job only counts as done when there's evidence it happened."
The comments come as businesses face growing pressure to put controls around autonomous software agents. Beard pointed to a gap between the separate governance layers available to large companies and what smaller firms need when deploying their first agent.
"A large enterprise can buy a governance layer," said Beard. "A smaller business running its first agent, however, needs those controls built into the agent from day one."
Evidence checks
A central part of the company's approach is a rule that a task is marked successful only when there is deterministic evidence of the result. That means either a process exits cleanly and creates the expected output, or an external system returns a real record ID.
Since August, its main write paths have returned a receipt stored in a ledger. That ledger is then checked against the external provider four times a day.
"Models will report success, pretty confidently, when they've failed," said Beard. "That's why a write now either produces a receipt or we treat it as not having happened."
Single send gate
The company has also placed every outbound message behind what it describes as a single send gate. It said the gate runs nine checks covering approval, approver, payload, mode switch, send whitelist, recipient policy, budget, suppression list and transport.
No message is sent unless a person has recorded approval. Approved content is fingerprinted and checked again before release so it cannot change after review.
Audit trail
Another control is a write census, which ADI described as a list of every point in the code where the agent can alter something in an external system. The list is generated by scanning the codebase, and a test fails if the list and the code do not match.
That process exposed gaps in an earlier version. Beard said the first census missed seven files because it recognised only one of two common code shapes.
One of those seven files could create, change and delete calendar events without any record being kept. The issue illustrates the kind of low-visibility risk that can emerge when agents are connected to day-to-day business tools.
"Nobody would have noticed until a meeting went missing," said Beard. "You can't log only the important ones and call it an audit trail."
Hidden failures
The company said its work on operational controls also uncovered problems that looked like successful runs. In one case, 17 scheduled jobs were configured so that a crash counted as a successful exit, meaning the failure alert never triggered.
ADI also found an inbox watcher that used Gmail's unread flag to determine what it had already seen. If a reply was first read on a phone, the message disappeared from the watcher's view and was not processed again.
Over one seven-day period, 86% of fresh model tokens were consumed by a background heartbeat that was enabled by default in the runtime and left unconfigured. Beard said the discovery led to a stricter default approach in which any missing setting that could affect a person-facing action is turned off.
"Now if a setting's missing and it could reach a person, it's off," said Beard.
Permission tiers
ADI said the agent's permissions are tied to a scoring system based on its own run log. Clean runs with evidence earn points, while failures and incidents reduce the score.
Once the record qualifies, the scoring system can flag the agent as eligible for a new permission tier. A person must still approve any move to a higher tier, and only a human can make that change.
The firm said it applies the same structure to client work, including the approval gate, evidence-based success tests and write ledger. Each client's names, figures and operating rules remain separated within that client's deployment.
ADI works with businesses in Australia and the United Kingdom, focusing on installing and maintaining AI agents inside existing workplace tools. Beard said that for many businesses, the central issue is no longer whether a model can generate text, but whether the organisation can trace what the software was allowed to do and what it changed.
"Look, the model is the easy part now," said Beard. "What a business actually needs to know is what the agent's allowed to do and who said yes, and then what it changed."