Nobody would sign for it
A fluent answer is not a correct one. When an assistant cannot show its sources, its assumptions and what it does not know, the board keeps it away from anything that matters. So it stays a toy.
From the RJ·AJ lab in Riyadh
AI that prepares the decision. You keep the veto.
Built for companies that want automation without giving up authority. Anything consequential waits for a named person to sign.
Try the gateno tier yet
Pick a request. The gate decides what runs alone and what waits for a name.
- ATLASlocal system, trades alone inside limits
- BOSLAlocal agent, corporate strategy unit
- SANADlocal agent, HR in pilot
- O24Uwebsite, luxury retail online
- AIPICKSCOREweb service, AI visibility, live
Why most AI deployments stall
The limits were never written
Capability is easy to add. Restraint has to be designed. Most projects start with what the system can do and discover, too late, what it must never do.
The data left the building
Employee records, deal flow and customer files get pasted into someone else's cloud. The most secure record is the one that was never made, and the second best is the one that never left.
How it works
- Weeks 1–2
Write the limits before the code
We sit with the person who will sign and decide what the system may never do. That refusal becomes a separate mechanism the rest of the system cannot reach or switch off.
- Weeks 3–6
Build the gates
Independent checks between the model and the world: sources verified, evidence attached, risk tiered, every step recorded. The model argues the case. It cannot act.
- Months 2–3
Prove it on paper
The system runs in your company, on your data, against real work, without touching anything real. Its record is inspected by your people, not ours, until it earns the next tier.
- From month 3
Hand over the authority, keep the veto
Reversible chores run alone. Consequential steps wait for a named signature. A kill switch outside the agents stops everything. In both languages, plainly.
What we build
Brief
A strategy office for leadership
Watches your approved sources, reduces them to the few signals that deserve attention, and prepares options before the meeting rather than after it. Six auditable checks on every answer.
- For
- CEOs, boards, strategy and investment teams
- Never
- Fills a gap in evidence with fluent language
- Built from
- BOSLA
Desk
A specialist that serves a whole workforce
An HR, operations or support specialist that answers in Saudi Arabic and formal English, drafts documents, tracks deadlines against the Kingdom's rules, and escalates the sensitive categories to a named approver.
- For
- Companies with a small team serving many people
- Never
- Lets an employee record leave your machines
- Built from
- SANAD
Gate
Governance for AI you already use
A rule engine, evidence ledger and human sign-off layer wrapped around the assistants and agents your teams have already adopted. Risk tiers decide what runs alone and what waits for a name.
- For
- Teams that moved fast and now need to be able to prove it
- Never
- Allows a bypass. Zero, by design
- Built from
- The ATLAS gates
The rules we work by
- Rules before code. What the system may never do is decided first and enforced by a mechanism it cannot reach.
- Evidence, not confidence. Every output carries its sources, its assumptions and a plain statement of what remains uncertain.
- Private by default. Data stays where it is created. Models run on your machines wherever they can.
- Paper before real. Every system earns its authority slowly, against a record anyone can inspect.
- Human authority. The consequential step waits for a name. It always says it is an AI.
Proof, not promises
Veto is the service form of what the RJ·AJ lab already runs with its own money, staff and name. Only ATLAS acts alone, and only inside limits a person wrote first.
Local systemATLASA trading system that runs on its own, day and night, inside hard limits: trades per day, a daily loss stop, a Sharia screen. A month on paper first.
Local agentBOSLAAn agent for executive briefing through six auditable checks. Automation earned by risk tier. Kill switch outside the agents.
Local agentSANADAn HR agent in pilot at a Saudi company. Arabic and English. Four categories always need a named approver.
WebsiteO24UA luxury automobile retail website in Riyadh. Verify identity, sign, pay and receive ownership online. Money before metal.
What the waitlist gets
A fixed-scope pilot
One workflow, one named approver, one paper phase. A written limit set before a line of code. A price agreed before the work starts.
Both founders in the room
We take a small number of companies per quarter because the two of us do the work. One finds the reaction. The other contains it.
Everything stays yours
Your data, your machines, your rules, your ledger. If we part ways, the system keeps running and the record stays readable.
Questions we get
Is this a chatbot?
No. A chatbot answers. Veto prepares a decision, attaches the evidence, and waits for a person. The model can argue; it cannot act.
Where does our data go?
Nowhere, by default. We run models on your machines wherever we can and collect nothing we do not need. Where a cloud model is unavoidable, that is a written, signed exception with a limit on what may pass through it.
Which languages?
Saudi Arabic and formal English, in the product and in the paperwork. If a system cannot explain itself to the person using it, in their language, it is not finished.
How long until something real runs?
Usually a quarter. Limits in the first two weeks, gates by week six, then a paper phase against real work. Authority is handed over tier by tier, never all at once.
Who is behind it?
RJ and AJ, a two-person lab in Riyadh. AJ finds the opening. RJ turns it into a system that can be trusted. The five systems on the lab site are our track record.
Prepare the decision. Keep the authority.
Join the waitlist and we will write to you when your cohort opens.
Join the waitlist