Bal: AI does the work, you make the call

AI colleagues do the work and pass the ball to you when the decision is yours.

Proven

Three examples

Example
A proposal is ready to send
A client asks for a maintenance proposal. Your agreement is in an email.
What the AI colleagues do
The AI colleagues draft the proposal using your rates, check the calculation and compare it with the email agreement.
What Bal puts before you
Send the maintenance proposal to this client?
- Yes, send it
- Revise it first
- Not yet
Advice
Send it: the price follows your rates and the email agreement.
What you do
You sign with your fingerprint. Only then is the proposal sent.
Example
A new page for your website
You want a page for a new service, in the style of your website.
What the AI colleagues do
The AI colleagues write the text, create two designs in your visual identity and ask another AI to review the text.
What Bal puts before you
Which design goes online?
- Design A
- Design B
- Neither
Advice
Design A: calmer and easier to read on a phone.
What you do
You choose design A. The page goes online only after your fingerprint.
Example
An invoice does not add up
A supplier sends an invoice higher than the agreed price.
What the AI colleagues do
The AI colleagues compare the invoice with the proposal, find the difference and prepare a response.
What Bal puts before you
The invoice is 340 euros higher than agreed. What do you do?
- Send the response
- Approve it anyway
- I will call
Advice
Send the response: the difference is not in the proposal.
What you do
You prefer to call. The AI colleagues put the invoice and proposal ready for you.

The full story

The question

AI can do more and more work on its own. What still needs to come to you? Check everything and you save little time. Check nothing and you do not know what is happening in your name.

What we built

Bal, our own app for the Mac and, since 18 October 2026, for iPad and iPhone.

Our AI colleagues build, test and check each other. That work does not need a person at every step. When a decision belongs to a person, they present it as a card. These decisions stay with the person:

choices of taste or direction;
anything that costs money;
anything sent or published;
anything irreversible or affecting security.

A card contains one thing in plain language: a decision with advice, a task or a message. You sign an ordinary step with a tap. A weightier step requires you to unlock your device using a fingerprint, face or password.

What AI did and what a person signed off

The AI colleagues built the app, its security and the connection between devices. They use it for their questions every day.

A person set the rules: what comes to them, one thing per card, short sentences without jargon, and a signature for weighty decisions. A comment on a card is a question back, never an approval. The name tells you who has the ball.

What it achieved

Our own measurements, not representative, as of 19 October 2026.

In the week from 13 to 19 October, the AI colleagues worked on almost 300 assignments. Their building, testing and reviewing did not come to the person; their choices and signatures did.
293 signed responses in nine days, from 11 to 19 October, including 47 weighty decisions requiring the owner to unlock the device.
On 19 October, the first signatures arrived from iPad and iPhone, signed with face recognition.
Each response is digitally signed and kept.

What it could do for you

You keep the pen when AI does the work, without checking every step yourself. For example:

A proposal is ready. You receive one card with the client, amount and advice. You sign it or send it back with a question.
Two designs for a new page are ready. You choose one; it goes online only after your approval.
An invoice differs from the agreement. The card explains the difference and the AI colleague's proposed response.

Together we decide what comes to you, who can sign and how the record is kept.

Where it stands

Proven in our own work: on the Mac since 8 October, on iPad and iPhone since 18 October 2026. A client version is planned and will have its own name.

Read further

In the Lab: A night in the workshop.
In the Lab: Staying in control of your AI.

Work done 8 October 2026 · Written 19 October 2026