People + AI Agents: How AppWT Runs Its Own Operations

AI Agents Business Automation Human Oversight

Why AppWT Is Its Own Case Study

Most of AppWT's client work is confidential. AppWT's own operation is not, so it is the plainest proof of how people and AI agents can share the work of a business. Everything on this page comes from AppWT's own operating records and its public team page.

Who Does the Work

• A core team of 7 people. The founder handles strategy, product direction, architecture and quality standards. The Team Supervisor handles team coordination and consistent execution. Specialists do and validate the work in their own field.
• 42 named Digital units. These are software AI agents. Each one has a name and a defined role, and each one is labeled as Digital on the team page. They are software, not staff, and they hold no management authority.
• Seven departments, six units each. The departments are Digital Solutions and Innovation, Guest Experience, Sales and Marketing, Security and Safety, Guidance and Education, Research and Development, and Reception and Mobility.
• Defined roles. Unit roles include patch and update work, audit and compliance reviews, testing, API integrations, business automation, prospecting and aftercare follow-up.
• Vetted independent contractors. The founder assigns specialist tasks to a vetted network of independent contractors, most of them in the USA.

How People and Agents Split the Work

The split is simple. The agents do the defined, repeatable work, and people make the decisions.

• Every task has a named doer. Work done by one unit is checked by another unit before it counts.
• Independent tasks run at the same time. Work that does not depend on other work starts right away instead of waiting in a queue, and two units never edit the same file at once.
• Some decisions always go to an owner. Deleting anything, handling a credential, setting a price, making a claim only an owner can make and touching anything on the protected list all wait for an owner's word.
• Owner decisions are written down with a date. The checks read the current ruling, so an agent follows today's rule and not an old one.
• A person approves before a client sees it. Human specialists review the units' work, and the founder gives final approval before any of it reaches a client.

The Rules Every Agent Works Under

1. A backup before every edit. A change to an existing file is refused until that file can be restored. The backup is test-restored at the time of the change, and a restore brings back the file's owner and permissions along with its contents.
2. Two matching checks before anything is called done. The two most recent runs of a check must give the same result. A check that could never fail does not count, and the test that could disprove a claim runs first.
3. Check what the customer sees. A website change counts as shipped when the live page proves it in a real browser, not when a file is saved.
4. Nothing is removed without an owner's direct instruction. Adding and restoring are allowed. Taking something away, such as an email forward, a DNS record, a scheduled job or a user account, needs an owner to name it first.
5. Guards that refuse risky changes. Software checks stop risky commands before they run. Examples include a permission change that sweeps across many files or into system folders, and any write to a protected file. A trial run is the default for bulk changes.
6. A refusal is reported, never routed around. When a guard blocks a change, the agent leaves the file as it was and reports the change it wanted to make. It does not make the same change another way, and no one loosens a guard just to make a feature work.
7. Every guard is tested both ways. Each guard has a case it must block and a lookalike it must allow. That keeps it from missing a real risk or blocking ordinary work.
8. "Unverified" is an honest answer. If a check cannot run, the result is reported as unverified, never as a pass. A tool that fails with an error has not shown that everything is fine.
9. Every change is logged. Each change on each website gets a dated entry in that site's log on the same day.
10. Alerts report what changed. Routine monitors log and flag their findings instead of filling an inbox. A finding is news when it is new, and one nobody acknowledges is raised again at 7, 14 and 28 days.
11. Every fix stays watched. Each fix is recorded as a lesson, added to the daily audit and a silent-failure audit, and written up in plain language for the person who reads the session audit.

This Page Was Made the Same Way

A Digital unit drafted this page from AppWT's own records. Before the page went live, the index file it changed was backed up and test-restored, and the page passed two matching checks in a real browser. The work was then logged for review.

Measured Facts

Honest Status

This page describes how AppWT runs its own work. It is not a client result, and it does not promise a result for any business. The people and unit counts match the public team page as of October 31, 2026.

Most of these rules were added after a real problem, so the same problem gets caught the next time. No set of rules removes every mistake, which is why a person still makes the final call.

What This Means for Your Business

If your team repeats the same tasks every week, AI agents can take on the defined parts while your people keep the decisions. AppWT can set them up with the same rules it uses on its own work: clear roles, a backup before every change, two matching checks, a log of every change and a person who approves. See enterprise automation and AI systems, the personal AI agent blueprint, AI consulting and the AppWT team.