Nearly every business product now has a sparkle icon somewhere. That tells you very little. A chat box bolted onto a project tool and a system that prepares your invoice reminders both get sold as AI-powered, and they carry very different risks and benefits.
You're right to be sceptical. This page breaks the label into four verbs, gives you questions to put to any vendor, and lists the red flags. Then it answers the same questions about Nautis, which is our product, including the places where the answer isn't flattering.
Four things AI can actually do here
Ignore the words "assistant", "copilot" and "agent", because vendors use them interchangeably. Ask which of these four verbs the product performs, on which records. Each one needs more trust than the last.
| Verb | Real version | Label-only version | Five-minute test |
|---|---|---|---|
| Reads | Answers from your live records, and shows which one | Answers from whatever you pasted into the chat | Ask about a deal you changed an hour ago |
| Ranks | Says what matters today, across functions, with a reason | Sorts one list by due date | Ask why item one is first. Open the record behind the answer |
| Drafts | Prepares the reminder, update or note using your facts | Generic text with your company name swapped in | Compare the draft with what you'd have sent |
| Acts | Changes a record or sends something, after a rule or a person allows it | Tells you to go and do it | Find what stops it, and what you can undo afterwards |
Drafting is the cheapest verb to add, so it's what most add-ons ship first. That doesn't make it worthless. If your data already sits in one good system, drafting inside it may be all you need, a case our guide on AI native vs AI add-on works through. Ranking across functions is rarer, and acting is where the risk lives.
Questions on data access
Start here, because the answer shapes everything else. An agent's usefulness and its risk both come from what it can see.
- Which sources does it read, by name? "All your data" isn't an answer. Ask for the list, and ask which of them are live.
- Does it copy data or fetch it on demand? A mirrored copy is another place your data lives. Ask how long it stays there.
- Does it respect each person's permissions? An employee shouldn't be able to ask the AI about salaries they can't open themselves.
- Is our data used to train models? Get this in writing. Notion, for example, says on its AI page that it won't use your data to train its models unless you opt in, and that contracts bar its AI subprocessors from doing so. Its pricing page lists zero data retention with LLM providers on Enterprise.
- Which model providers are involved? If you handle client data under contract, you may need to tell your clients.
Questions on approvals and audit logs
A published sentence about human oversight isn't enough. Ask to see the screen. Lindy's pricing page, for instance, says anything with outside impact waits for approval, such as sending emails or publishing docs, while read-only lookups from approved sources don't need it. That's the kind of clear rule you want a vendor to commit to. The sorting of which actions should wait is covered in human in the loop AI agents.
- What's the default? Outbound actions should default to asking. If autonomy is on from the first day, walk away.
- How specific is the control? Per action or capability is useful. One master switch tends to end up either fully on or fully off.
- What does the approval show? The exact recipients and full text, not "send follow-up email".
- Who may approve? A named role. If anyone can click yes, you have a formality and no control.
- What gets logged, and for how long? Ask whether the log covers actions that ran automatically, who approved what, and what the agent based its output on.
- Can you export the log? You may need it for a client or an auditor. Ask before you depend on it.
What does an AI-powered business management platform cost in credits?
More than the seat price suggests, once you use it properly. Vendors meter AI in units that aren't comparable, so don't compare the allowances directly. Ask what one real job costs, then multiply. These are published figures as of November 2026.
| Vendor | Unit | What's published | What to check |
|---|---|---|---|
| ClickUp | Credits, pooled per user | Core 1K, Business 1.5K, Enterprise 3K a user a month. Extra: $10 per 10K | How many credits your agents burn on a real task |
| Notion | Notion credits | Notion Agent on Business at $20 a member. Custom Agents: $10 per 1,000 monthly credits after a trial | Whether you need Custom Agents, or Notion Agent is enough |
| Lindy | Credits | Team from $29.99 for 3,000. Everyday asks 2 to 250 credits, deep work 250 to 1,000, big builds 1,000 to 2,500 | That range is wide, so measure your own tasks |
| HubSpot | HubSpot Credits | Smart CRM includes 3,000 on Professional, 5,000 on Enterprise | What each AI action costs in credits |
| Nautis | AI credits per workspace | 1,000, 2,000 or 2,600 a month by plan. Sending is blocked at zero until reset or top-up | Whether your routines fit the plan's allowance |
Odoo takes another route: its pricing page lists Agentic AI within its Custom plan, not its Standard plan. So a feature on the marketing page may belong to a higher tier. Ask which plan your use case needs, and what happens at zero: does it stop, or does it bill you? Neither answer is wrong. You just need to know before an agent runs overnight.
Red flags
- No named data sources. If the sales engineer can't list them, assume the demo was staged on sample data.
- Demo only. No trial, or a trial where you can't connect your real records.
- Time-saved claims without a method. Ask how the number was measured. If it came from a survey of the vendor's fans, set it aside.
- Full autonomy as the headline. Outbound email to customers with nobody checking should worry you, however well the demo reads.
- A log you can't open. If the only record is a chat thread, you can't reconstruct what happened last Tuesday.
- Unexplained metering. A price you can't estimate for a normal month will surprise you eventually.
- AI priced in, value undefined. If the price rose and you can't name the job it now does, you're funding the vendor's roadmap.
One flag isn't fatal, since young products often have rough edges. Two or more together usually means the AI is a sales feature, and you can buy the rest cheaper without it.
Nautis, on those same questions
Nautis is the AI business operating system for startups, and its Chief of Staff works on top of its modules. Here's how it answers, based on its own help center. It launched publicly in July 2026, so few independent reviews exist, and you should verify this during the 14-day trial.
| Question | Answer | The limit |
|---|---|---|
| What does it read? | Workspace records. The Briefing scans tasks, pipeline, runway, meetings, approvals, overdue invoices and more, and skips sources with no data | A thin workspace gives a thin briefing |
| Copy or on demand? | Connected apps aren't mirrored. Agents call actions on demand | Nothing from them is stored in Nautis, so an agent has to call the tool to see it |
| What runs without asking? | Capabilities default to Ask every time, set per capability under Agent Autonomy. Quick capture to a private inbox runs on its own | Admins can switch to Run without asking, and a delete in a connected app then counts as a write |
| Who approves? | Owners and admins, in the Approval Center | Someone has to review, so it isn't hands-off |
| What's logged? | Writes, approvals and background runs, in AI Activity | Read-only chat lookups aren't listed. Owners and admins only, with no export on that screen |
| Training on our data? | The help center doesn't state a policy | Ask before you put sensitive data in |
| Cost? | $49, $99 or $129 a month, with 1,000, 2,000 or 2,600 credits, as of November 2026 | No free version after the trial. Credits are capped, with paid top-ups |
The details are in the AI Governance guide, and the morning workflow is covered on the AI Chief of Staff page. Nautis is broad, not deep: a dedicated tool will often go further in CRM, accounting or project management. It's also not an accounting system or a payroll provider. If your data already lives happily in one strong system, its add-on AI may serve you better. See Nautis vs Notion and Nautis vs ClickUp for two cases.
A one-hour scepticism test
You can run this in any trial, and it doesn't need a sales call. Load real data from the last two weeks first, since an empty workspace proves nothing. Then do these in order, with a notepad for what surprised you.
- Ask a question whose answer you already know, about a customer or a deal, and check it against the record.
- Change one record, wait a few minutes, and ask again. A stale answer tells you how fresh the data really is.
- Ask it to draft a message to a real contact. Count how many facts you'd need to fix.
- Tell it to send that message. Note exactly what stops it, and who gets asked.
- Find the log entry for what you just did. If you can't find it in two minutes, an auditor won't either.
- Open the usage screen and note what that hour cost. Multiply by your real working pattern, and you have a monthly estimate no sales page will give you.
FAQ
What is an AI-powered business management platform?
It's business software where AI works on your company's own records: it reads them, ranks what matters, drafts next steps and can act with permission. Many products labelled this way only draft text. Ask which of those four things it does, and on which records.
Is AI business software safe to use with customer data?
That depends on the vendor's data terms and your own contracts. Ask which model providers process the data, whether it's used for training, and how long it's kept, and get the answers in writing. Check your client contracts too, since some restrict where data may go.
How much do AI credits cost in business software?
There's no standard, because each vendor defines its own unit. As of November 2026, ClickUp sold add-on credits at $10 per 10K, while Lindy charged $10 per 1,000 extra and Notion $10 per 1,000 for Custom Agents. The credits aren't comparable, so measure what one real task costs.
Does Nautis let AI act without approval?
Not by default. Capabilities start on Ask every time, and owners or admins can change them per capability. Deleting Nautis records always asks. Quick capture to a private inbox runs on its own, and every action, automatic or approved, is audit-logged.