What’s our liability cap in the Tellbrook Freight MSA, and are data breaches carved out?
The top stone · Agents for the whole company
Give everyone AI. Decide what it reaches.
People bring their own AI to work when the approved one is worse or rationed. Pebbl AI gives everyone agents that answer from your company’s own knowledge and work inside the systems underneath, with no seats to hand out. Your teams own the knowledge. Your admins decide what each agent reads, which model it runs on and the rules it follows.
Illustrative · Invented company and data
Why people go around it
Control that doesn’t start with no.
People don’t reach for their own AI to cause trouble. They do it because the approved tool can’t see the company’s documents, runs one vendor’s models, or is rationed: a seat for some, a usage window for the rest. Block the workaround and the work tends to move to phones.
The way back is an approved tool that does the job better than the workaround, with your controls built into it rather than bolted on.
EY’s 2025 Work Reimagined Survey found that, depending on sector, 23% to 58% of employees bring their own AI tools to work.1
One in five respondents say their organization is limiting AI use because of operating costs, including token costs (McKinsey, 2026).2
Half of regular AI users surveyed say their company lacks clear guidance on managing teams of people and AI agents; 47% rank AI-driven accountability a top-three concern (BCG, 2026).3
Sources
- 1 EY, “EY survey reveals companies are missing out on up to 40% of AI productivity gains due to gaps in talent strategy” (2025 Work Reimagined Survey), November 10, 2025
- 2 McKinsey & Company, “The state of AI in 2026: On the road to ROI”, August 25, 2026
- 3 Boston Consulting Group, “AI at Work: Strategy Matters More Than Tools”, June 2026
It reaches the work
It answers. Then it does the work.
Most agents can read a document and describe it. Pebbl AI’s agents also work inside Pebbl’s own products, with each person’s own permissions, and anything that can’t be undone waits for a yes.
A budget owner submits through the agent.
A designer logs the day in a sentence.
Pebbl AI is the top stone. PebblPlan and PebblTime are the systems that hold the work, on one platform with one sign-in, one admin console and one bill. See how the suite stacks.
Illustrative · Invented company and data
One day at Halden
One company. Four teams. One working day.
Halden is a distributor we made up, with 1,400 people at sites in Leeds, Columbus and Rotterdam. Here is a Tuesday. Each agent below was set up by Halden’s admins for one job, and reads only the libraries it was given.
08:52 · People · HR policies
A policy answer, with the paragraph.
A site manager is flying to Rotterdam on Sunday for a Monday audit. He asks HR policies whether he can book premium economy, and whether he gets the Sunday back. The agent searches the handbook, opens the working-time policy and answers both in four sentences, with a number after each claim. Hover a number and the passage opens, with the document and the page.
Illustrative · Invented company and data
Halden’s HR policies agent answers a travel question in four sentences with numbered citations; the second opens the working-time policy at page 7. 10:05 · Sales · Pipeline review
This week’s export, as a chart.
A sales lead attaches the week’s pipeline export for Pipeline review and asks where the quarter’s commit moved, and which deals slipped. She gets the headline figures, a chart by stage against last week, and a table of the slipped deals, checked against the stage rules in Halden’s sales playbook. She downloads the table as a CSV for her one-to-ones.
Illustrative · Invented company and data
Halden’s Pipeline review agent reads a weekly CSV export and answers with three headline figures, a bar chart of the pipeline by stage for two weeks, and a table of four deals that slipped into the next quarter. 11:20 · Support · Returns desk
The process, drawn from the policy.
A new starter on the support desk asks Returns desk how a large return is handled. Instead of a paragraph, she gets the process as a flowchart drawn from the returns policy, with credit control’s sign-off as its own step and the policy cited underneath. She can switch the drawing to its source, or download it as SVG for the team’s wiki.
Illustrative · Invented company and data
Halden’s Returns desk agent answers with a flowchart of the returns process drawn from the returns policy, with credit control’s approval as its own step. 14:30 · Finance · Month-end close
Two spreadsheets in. Variances and files out.
The controller hands Month-end close September’s trial balance and the opex budget, and asks for every cost center more than 5% off budget, with a first draft of the commentary for Thursday’s ops review. The agent works through both workbooks in an isolated sandbox with no internet access, returns the variances as a table, and hands back a workbook and a Word draft laid out the way Halden writes board memos. Code execution is on for this agent; it’s off for every agent until an admin turns it on.
Illustrative · Invented company and data
Halden’s Month-end close agent takes a trial balance and a budget workbook, returns a table of seven cost centers more than 5% off budget, and hands back a workbook and a Word draft.
Nobody in this story bought a seat or ran into a personal usage window. Each of them picked, by name, an agent set up for that job.
Shaped by change
Rules change. Every agent keeps up.
New policies, new teams, better models, new people. You decide each one once, in one place, and every agent works from it the next time anyone asks.
- A new policy
Write it once. Every agent follows. Your tone, what to refuse and what to escalate sit ahead of every agent’s own instructions.
- A new team’s knowledge
Its owners decide who reads it. Before a library reaches an agent, Pebbl AI shows how many people, roles and documents that means.
- A better model
Choose it agent by agent. Each agent runs on the model and thinking level you set, once it has passed testing, on your provider keys or ours.
- A new hire, a contractor
Your identity provider decides who gets in. Require single sign-on, with break-glass exemptions that carry a reason.
Illustrative · Invented company and data
Sign-in and privacy
Your sign-in. Their conversations.
People sign in through your identity provider over OpenID Connect, with a passkey, or with a code sent to their work email. There are no passwords. Your admins can require single sign-on, set how long a sign-in lasts, and keep break-glass exemptions with a reason and an optional end date.
Conversations belong to the person who had them. There’s no screen, for your admins or for ours, that opens someone else’s.
Illustrative · Invented company and data
No seats
No seats to ration.
Invite the whole company. You pay for the people who use it each month, up to a ceiling agreed up front, and anyone who doesn’t use it in a month doesn’t count that month. If adoption runs below plan, the bill falls with it, as a credit line on the invoice. On your own provider keys, your provider bills the tokens; on ours, they’re on the same invoice at published provider rates. The billing page shows what each agent cost and who was active.
Illustrative · Invented company and data
Bring us your AI policy.
Send us the policy you wrote for AI at work. We’ll walk your IT and security teams through where each rule would live in Pebbl AI: who reads which knowledge, which models answer, what every agent follows, and who gets in.