Use cases
What AI does for your team, and what a person still approves.
Find your team below. Each card gives three lines from a scenario written up in full on the blog: how the work runs today, what AI takes on once the business objects, rules and tools are defined, and which actions stay with a person — with permissions and audit enforced by the runtime, not by the prompt.
- 6 teams
- Sales, service, finance & procurement, HR, legal & compliance, projects
- 10 scenarios
- Each written up in full on the blog, in eight languages
- 7 templates
- Forkable starters behind seven of the scenarios
What each scenario needs before it runs: Works on the platform todayNeeds a connection to a system you runNeeds the business rule written down first
How to read a card
Three lines, the same in every scenario
The cards are short on purpose. The full write-up behind each one shows the objects, the rules and the first build; the template, where there is one, is the app you can install or fork.
- Today: where the work stalls now, and why a form or a queue is not enough.
- With AI: what the app does once the work is modelled — read, classify, draft, check, propose.
- A person approves: what stays a human decision, and what the runtime enforces around it.
- Agents act as the signed-in user, inside that user’s permissions; the records they read and change are audited, and on Enterprise so is each tool call the in-app AI makes.
Sales
Keep the CRM current from the conversation
The selling happens in meetings, emails and notes. The record should follow from them, not wait for a rep to type it in.
AI sales assistant
Needs a connection to a system you run
- Today
- Reps fill in CRM fields after the selling is done, while the real context sits in meetings, emails, notes and follow-up promises.
- With AI
- Reps update by voice or text. AI summarizes meetings, picks out next steps, surfaces deal risk, suggests stage changes and preps the next account meeting.
- A person approves
- The rep confirms each suggested CRM change, and customer-facing commitments need confirmation. Managers set the risk rules and approval boundaries.
Customer service
Read the customer’s problem before it waits in a queue
Support work is language, judgment and repeated decisions: a good fit for AI that drafts and a person who decides what is sent.
AI ticket hub
Needs a connection to a system you run
- Today
- A long email or a frustrated chat becomes a record that waits for someone to read, classify, search, reply and escalate.
- With AI
- AI classifies each issue, detects urgency and sentiment, matches knowledge articles, drafts replies with citations and ranks cases by SLA.
- A person approves
- Agents confirm drafts, routing and case updates. Refund, compensation and contract-promise replies can be set to wait for a manager’s approval; escalations are created only after permission checks.
Finance & procurement
Judge spend with the evidence in front of you
Recognising an invoice is the easy part. The decision needs policy, budget, history and an approval chain that can explain itself.
AI expense audit
Needs a connection to a system you run
- Today
- OCR fills in the fields, but finance still judges by hand whether a claim is in policy, in budget, project-related and not a duplicate.
- With AI
- AI extracts receipt data, checks policy, budget, project and duplicate invoices, ranks the claims to review first and cites the policy line behind each finding.
- A person approves
- AI recommends; payment approval follows a clear authority chain, high-risk claims route to finance review, and reviewer corrections are recorded.
AI procurement risk
Needs a connection to a system you run
- Today
- Supplier risk is spread across systems: expiring certificates, slipping delivery, unusual pricing, contract gaps and quality returns.
- With AI
- Before an order is placed, AI checks qualifications and delivery, compares the quote with history, surfaces quality and contract issues and suggests alternatives.
- A person approves
- AI does not make the buying decision: procurement approves with the risk explanation in front of it, and review outcomes feed back into supplier governance.
HR & employee services
Let employees ask, then run the right process
Employees do not think in system categories. The service center should understand the request first and then route it safely.
AI employee service center
Needs the business rule written down first
- Today
- Employees need a VPN, a laptop, leave or a policy answer, and the portal makes them find the right HR, IT, admin or finance form first.
- With AI
- AI splits one message into the right service requests, pre-fills the employee’s context, answers policy questions from the knowledge base and routes fulfilment tasks.
- A person approves
- Requests are created only through a governed service catalog; access, devices and reimbursements stay behind permissions and approvals, tracked in the same conversation.
Legal & compliance
Flag the risk first; the responsible owner signs
Contracts and policies are dense text. AI can read them first and draft findings; responsibility stays with legal and the control owners.
AI contract review
Needs the business rule written down first
- Today
- The contract register is managed but the review is not: risk stays in attachments and comments, and approvers see too little context.
- With AI
- AI extracts parties, amounts, terms and non-standard clauses, compares them with the standard template and turns deviations into risk records.
- A person approves
- Legal confirms or edits every finding and owns the final opinion. High-risk findings go to approval, and signed obligations become follow-up tasks.
AI compliance and internal controls
Needs a connection to a system you run
- Today
- Compliance teams align policy text with evidence scattered across IAM, contracts, procurement, logs, workflows and spreadsheets — by hand.
- With AI
- AI parses policies into controls, matches evidence, explains why a gap is high risk and suggests remediation tasks.
- A person approves
- Responsible owners confirm high-risk findings and evidence requirements; remediation and its confirmation stay in the audit history.
Projects
See the delay in the status update, not the missed date
Project risk rarely announces itself. It shows up as small comments long before a milestone slips.
AI project management assistant
Needs the business rule written down first
- Today
- Risk hides in meeting notes, task updates, weekly reports, change requests and resource conversations until a milestone slips.
- With AI
- AI links notes to tasks and milestones, detects blockers, dependency and resource conflicts, creates risk records with suggested actions and answers health questions.
- A person approves
- AI detects and explains; commitments stay human decisions. Managers turn accepted risks into action plans; a rule can hold scope changes until their impact is assessed.
Across teams
Two patterns any team can reuse
Not a team of their own: a content workflow and a general agent workbench, both under the same line between suggesting and doing.
AI content workbench
Needs a connection to a system you run
- Today
- AI may speed up one draft, but ideas, sources, review comments, the publishing calendar and retrospectives stay scattered.
- With AI
- AI clusters source material into topics, writes briefs with audience and evidence, highlights claims that lack support and summarizes performance after publishing.
- A person approves
- Drafts route to brand, product, legal or customer-success review before publishing, and review comments and approvals are structured records.
AI agent workbench
Needs the business rule written down first
- Today
- An assistant answers the question, then the user still opens the business system to create tasks, start approvals, update records and leave an audit trail.
- With AI
- The agent turns a goal into a visible plan and calls governed tools: it queries records, creates tasks, starts approvals and generates reports.
- A person approves
- It checks the user’s access before querying data, asks before creating tasks, holds high-risk actions at a flow approval step, and the records it reads and changes are audited.
The common line
What AI may do alone, and what waits for a person
| Kind of action | Examples from the scenarios | What happens |
|---|---|---|
| Runs on its own | Summaries, classification, drafts, internal reminders, risk flags, reporting | AI does it; the result is visible and recorded |
| Asks for confirmation | Creating tasks, status updates, routing, operational record changes | The person using the app confirms before anything is written |
| Needs approval | Customer-facing commitments, money movement, contract terms, access changes, data export, audit closure | An approver decides; the runtime does not run it otherwise |
FAQ
Frequently asked questions
Are these customer stories?
No. Each card summarizes a scenario written up on the blog: how a team’s work can be modelled and governed in an ObjectOS app. None of them reports on a named customer.
Do we have to start from a template?
No. Seven of the scenarios have a matching template you can install or fork; the others start from a description of the work. Either way the result is the same kind of app: objects, permissions, workflows and agent tools a person can review.
Who decides what AI may do on its own?
Your team does, in the app’s definition: which actions run on their own, which ask for confirmation and which need approval. The runtime enforces that line for people and agents alike, and agents act as the signed-in user, never as a privileged AI identity.
Next pages