FAQ and glossary

Questions about
governed agent workflows

What is SogaStudio?

SogaStudio LLC is a Valdosta, Georgia company that builds AI agent workflow systems. SOLODRIVE.PRO, a live software business, runs on its governed AI agent workflows, and SogaStudio builds the same kind of workflow for operators who already run a process by hand. For an operator, that means the repeat work in the business runs every day; releases, spending and new public channels wait for a person, and the operator decides what else does.

Is SogaStudio the game studio, or another studio with a similar name?

No. SogaStudio, written as one word, is SogaStudio LLC of Georgia, USA. It builds AI agent workflow systems, and SOLODRIVE.PRO, a live software business, runs on its agent workflows. It does not make games and is not connected with game, yoga, design or other studios that use a similar name.

What does a governed AI agent workflow do for my business?

It takes the repetitive jobs you already run by hand, such as intake, follow-up, scheduling, reporting, content or back-office checks, and runs them the same way every time. You get a status and a written result for every job, an alert when something breaks, and the final say on releases, spending, new public channels and which work may run without review. The method →

Will anything go out under my name without my approval?

Only the kinds of work you have approved to run on their own, and only after they pass automatic checks. Production releases, spending and new public channels wait for a person by default. You set where that line sits. Human gates →

What do I still have to do myself?

You make the calls that carry risk: production releases, spending, pricing, new public channels, and which kinds of work may run without your review. The research, drafting, building, testing and the repeat jobs in between are the agents' work.

What is a governed AI agent workflow?

A way of running work with AI agents where every task is a work-order record, each agent has one job and a write boundary backed by tooling, and a person approves production releases, spending and new public channels. Every run leaves a committed result that a reviewer reads before the work is closed. The result is agent work you can see, check and stop, instead of work you have to take on trust. How it works →

What is a work order in an AI agent system?

A single record for one unit of work. It names the agent that owns it, holds its status, cost and custody trail, and points to the result the agent wrote. A request that only exists in a chat or an email is not work until it is a work order. For you, it means you can see where every request stands, in one place.

Can the agents release to production on their own?

No. Agents build and test on staging. A person approves each production release, and the release tool shows a diff, backs up what it will touch, and cannot reach customer or payment records. Your customers see a release only after a person has reviewed it. Human gates →

Do the agents see passwords or API keys?

No credential value is shared in a message, a transcript or a result. Keys are kept in one store and named by the job they do. An agent asks for a key by job name and the value goes straight into the process that needs it. Your keys do not end up in chat logs or reports.

What happens when an agent gets stuck?

It measures the problem, names what kind of failure it is, raises an alert that repeats until a person acknowledges it, and marks the work blocked. It does not route around the problem or quietly deliver a weaker result. You hear about a problem when it happens, rather than finding a quiet workaround later.

Can an agent mark its own work as done?

No. An agent finishing its work marks it delivered. Done means a reviewer has read and re-checked the result. A run that commits nothing is recorded as blocked, whatever the agent reports. When your board says done, someone checked it.

Do you build this for other businesses?

Yes, for operators who already run a workflow by hand and can describe it step by step. We start from that workflow, automate one job at a time, and put human gates where money, customers or reputation are at risk. You keep the process you trust, and you take on change one job at a time. The method → · Tell us about yours →

Do I pay you, or do you take a share?

You choose. There are two ways to work with us. Pay for the work: you pay SogaStudio for the engagement. Or pay with a share: instead of a fee, your business gives SoGa Capital a percentage of its transactions going forward, and SogaStudio does the same work. SoGa Capital invests technology, not cash. Two ways to work with us →

What is SOLODRIVE.PRO?

A web platform that gives independent drivers their own booking page, their own repeat customers, reservations and recurring trips, and payment into their own Stripe account, alongside the ride-hail apps. It is the case study on this site: a live software business developed and operated on SogaStudio's agent system. solodrive.pro ↗

Where is SogaStudio based?

Valdosta, in Lowndes County, South Georgia, USA. SogaStudio builds AI agent workflow systems for operators in South Georgia and across the United States. It is not a storefront. Mailing address: SogaStudio LLC, 8735 Dunwoody Place Ste N, Atlanta, GA 30350 · (888) 609-1221.

Glossary

Work order
One record for one unit of work: owner, status, cost, custody and result. The single source of truth for what is being done.
Lane
One specialised agent with one kind of job, its own tool allowlist and its own write boundary.
Write boundary
The exact set of repositories and paths a lane may change. Anything outside it is refused before the write happens.
Gate
A step that waits for a person: production releases, spending, new public channels, rule changes.
Staging
A working copy of the product where changes are built and checked before release. Staging runs on test payment keys with outbound mail switched off.
Staged
A work-order state meaning the change is built on staging and waiting for a production release.
Delivered
The agent has finished and committed its result. Not yet reviewed.
Done
A reviewer has read and re-checked the result and closed the work.
Blocked
Something outside the agent has to change before the work can continue. The result says exactly what.
Signal, alert, decision
Types of message. Messages carry information and requests for decisions. They never assign work. Alerts repeat until acknowledged.
Durability check
The check that runs after an agent finishes. It confirms the declared result is committed, and records the run as blocked if it is not.
Negative control
A deliberate test that a check can fail. A check without one may be green because it cannot see the problem.