AI discovery → company agents
Your company already runs on AI. You just can't see it.
Sonsara finds the AI work your people already do — including the shadow AI nobody approved — and turns it into secure, compliant agents your company owns.
5 of 6 tools are outside your controls — company and customer data in personal accounts, with no audit trail.
| Tool | Used for | Risk |
|---|---|---|
| Finance | ||
| ChatGPT · personal | Reconciling supplier invoices | HighSupplier bank details in a personal account |
| Claude · personal | Month-end variance notes | MediumUnreleased financials, no audit trail |
| Support | ||
| Browser agent | Jira tickets from Slack threads | HighActs as an employee with their work token |
| Sales | ||
| Claude · own connectors | Follow-ups from call notes | MediumCustomer data through personal connectors |
| Gemini | Screening CVs | HighCandidate personal data in hiring decisions |
| Ops | ||
| Microsoft Copilot | Weekly status report | CoveredApproved and logged |
Requirements interview · Priya N., FinanceLive
Sonsara is listening — speak, or type below
Thanks, Priya. Walk me through it as you do it today — what's the first thing you open in the morning?
Xero, then the supplier statements in the AP inbox. I paste both into ChatGPT and ask it to match them up.
And when something doesn't match?
Small differences go straight to AP. Anything over £500 I check myself first.
Got it — over £500 comes to you before AP. Which suppliers cause the most mismatches?
Mostly freight, and the two packaging suppliers
Conversation agenda2/4 complete07:42
- ✓What you want built
- ✓How it works today
- The details we need
- Wrapping up
| Agent | Trigger | Model | Region | Runs | Spend |
|---|---|---|---|---|---|
| Finance agent | 66 | £2.10 | |||
| Supplier reconciliation New | Weekdays · 09:00 | Sonnet 4.6 | EU | 1 | £0.04 |
| Invoice Chaser | Every 6 hours | Haiku 4.5 | EU | 28 | £0.70 |
| Spend review | Xero · bill approved | Haiku 4.5 | EU | 37 | £1.36 |
| Sales agent | 412 | £18.40 | |||
| Renewal Radar Gong disconnectedReconnect | Daily · 09:00 | Sonnet 4.6 | EU | 35 | £1.90 |
| Call follow-ups | Gong · meeting ends | Haiku 4.5 | EU | 88 | £4.20 |
| Lead Enricher | HubSpot · new lead | Haiku 4.5 | EU | 97 | £0.90 |
| Support agent | 214 | £4.30 | |||
| Ticket Triage | Zendesk · new ticket | Haiku 4.5 | EU | 160 | £3.20 |
| Refund Checker | Zendesk · refund request | Haiku 4.5 | EU | 54 | £1.10 |
Today 6Kept 30 days · export any time
- SMDone · 41s
Sara Malik asked in Slack
Call follow-ups drafted 3 emails for Acme Ltd
Gong · readHubSpot · readGmail · draft
Data: Acme Ltd deal · call of 20 Oct
- Blocked
Policy · email outside Northwind
Spend review tried to email a supplier directly
Gmail · send → vendor.ioblocked
Owner Priya N. notified · approval step added at 10:20
- Done · 1m 12s
Schedule · weekdays 09:00
Supplier reconciliation checked 412 invoices, flagged 6
Xero · readOutlook · send → ap@PII · tagged
Data: 38 supplier statements · no data copied
- Failed
Schedule · daily 09:00
Renewal Radar could not start
Gong · refused
Owners Jamie R. and Lena B. asked to reconnect in Slack
- AMChange
Alex Morgan approved Supplier reconciliation v1
Change · into the vault, after a passing test run
- LBAudit
Lena Brandt exported the Q3 audit pack
SOC 2 · ISO 42001 · every run, change and access grant
Today
@Finance agent can you run supplier reconciliation for this week please
Supplier reconciliation is done. Two mismatches are over £500, so they've come to you first.
Ran Supplier reconciliation · run #1,284 · logged in Sonsara
Nice — this used to take Priya most of the morning 👏
2 replies
Can you run supplier reconciliation for this week?
09:14Supplier reconciliation · done
412 invoices checked. 6 mismatches — 2 are over £500, so they've come to you first.
Ran on Finance's Xero and Outlook · run #1,284 logged in Sonsara
Ask our Finance agent to run supplier reconciliation for this week and tell me what needs my attention
Used Finance agent · Northwind⌄
Done — Northwind's Finance agent ran supplier reconciliation on this week's statements.
It checked 412 invoices and found 6 mismatches. Two need you, because they're over £500:
- Halden Freight — invoice £1,240 higher than the statement
- Pack&Co — duplicate invoice for £860
The other four have gone to accounts payable. The run used Finance's Xero and Outlook accounts and is logged in Sonsara.
@Finance agent run supplier reconciliation for this week
Talked to Finance agent · Northwind
Supplier reconciliation is done for this week.
Invoices checked412
Mismatches6
Waiting for you2 (over £500)
Sent to AP4
It ran on Finance's Xero and Outlook accounts, under Northwind's controls, and the run is logged in Sonsara.
See every AI tool in use
Sonsara reads the activity your systems already log and shows which AI tools each team uses, what for, and what's at risk — including the ones nobody approved.
Detect AI usage across
The assistants, agents and automation tools your people already use.
- ChatGPT
- Claude
- Gemini
- Microsoft Copilot
- Perplexity
- Grok
- DeepSeek
- Mistral
- Meta AI
- GitHub Copilot
- Cursor
- Notion AI
- Zapier
- n8n
The problem
AI adoption didn't wait for a policy.
Your people found the tools before anyone signed them off. Personal ChatGPT accounts, Claude with their own connectors, browser agents posting in Slack on someone's behalf. Some of it is your best work. None of it is on the books.
01 · Visibility
You can't see it.
Which tools, on whose accounts, touching which data. Most of it never goes near an approved app, so it never shows up anywhere you look.
02 · Compliance
You can't audit it.
When a customer, an auditor or your SOC 2 assessor asks where your data goes, "we're not sure" is not an answer anyone wants to give.
03 · Continuity
You can't keep it.
The workflow lives in one person's account and one person's head. When they move on, the work walks out of the door with them.
A different answer
Don't block it. Adopt it.
Shutting shadow AI down pushes it further out of sight and throws away what your people have already worked out. Sonsara takes the other route: find the work, understand it, and rebuild it properly. Your people keep the speed. You get the control.
Block The usual security play
- Tools get banned, so the work moves somewhere you can see even less
- The time your people were saving disappears
- Nobody learns which work AI was actually good at
Adopt The Sonsara way
- Every tool, team and job mapped, with the risky ones flagged
- The valuable work rebuilt as tested, secure company agents
- People keep working the way they found works — just safely
Shadow agent vs company agent
Same task. Two very different agents.
The workflow is the easy part — anyone can build one in an afternoon. What turns it into something the business can run on is the seven layers around it.
Layer
Lives on one laptop, unknown to the company
Documented, controlled, audited, in the vault
01Specification
ShadowLives in one person's head and a prompt
CompanyWritten spec, documented, handover-ready
02Ownership
ShadowOne login. Gone the day they leave
CompanyOwner, editors and users, with roles and SSO
03Connectors
ShadowPersonal tokens pasted into the prompt
CompanyApproved connectors, scoped credentials
04Guardrails
ShadowNone. It will email a vendor if asked
CompanyPolicies block, tag or log actions. Model and region restricted
05Testing
ShadowTried once, then trusted
CompanyDry runs and simulations before go-live
06Audit trail
ShadowNo record of what it did or touched
CompanyEvery run logged: tool calls, data, outcome
07Compliance
ShadowInvisible to the CISO, the auditor and the insurer
CompanyEvidence pack for ISO 42001, SOC 2 and cyber insurance
What we do
We take that risk on. We're the MSP for agents.
Your staff build the agents. The ones that matter become company agents: registered, tested, owned, and run by us with a 30-minute fix.
01
Registered.
Every company agent is on the books: what it does, what it can reach, and who it belongs to.
02
Tested.
Each one passes test runs and checks before its owner approves it, and again after every change.
03
Owned.
A named owner signs it off. It runs on company accounts, so it stays when people leave.
04
Run by us.
We watch every run. If one breaks in production, it's ours to fix, within 30 minutes.
What's an MSP? A managed service provider is the outside IT firm most companies your size pay monthly to run their laptops, email and network. It doesn't build them. It keeps them working, and takes the call when something breaks. We do that for your agents.
How it works
From shadow AI to company agents, in four steps.
Discovery shows you where AI is already doing real work. Everything after that is about making that work something the company can own.
See every AI tool in use
We start from the activity your systems already record — Microsoft 365, Google Workspace, Slack, Jira and more — so the first picture needs nothing installed on anyone's laptop.
Find the jobs behind the usage
Patterns in that activity show what AI is actually doing: the reports it writes, the tickets it moves, the messages it sends on someone's behalf. Each one is checked for risk.
Ask the people doing the work
Where we find something worth keeping, we run short, targeted requirements interviews with the people behind it: how it works, what it touches, and where it breaks.
Rebuild it as a company agent
Tested before it runs, on company credentials, with every run logged. The owner approves a test run, then it goes live — and keeps running when people move on.
Company agents
The same work, done properly.
A company agent does the job someone was already doing with AI — but it belongs to the business, not to a personal account.
01Compliant
Every run is logged — what it did, what it touched, and who approved it. The audit trail exists before anyone asks for it.
02Secure
Runs on company-held credentials with only the access it needs. No personal accounts, no data going somewhere you didn't choose.
03Hardened
Tested on real cases before it goes live, watched while it runs, and fixed fast when something upstream changes.
04Owned
The know-how stays with the company. When someone leaves, their agent doesn't leave with them.
Supplier invoice reconciliation
finance · weekdays 09:00 · v3
- Requirements interview with the owner
- Tested against real invoices
- Runs on company credentials
- Every run logged
- Test run approved by the owner
Who it's for
Built for operations. Signed off by security.
Discovery gives both sides what they need from the same picture.
For the COO
Know where AI is already earning its keep, and scale the best of it.
- See which teams use AI, for what, and how often
- Pick the agents to build from real work, not guesswork
- Roll one person's shortcut out to the whole team
For the CISO and IT
A clear view of unsanctioned AI, and a safe place to move it to.
- Every tool and personal account outside the approved stack
- Which systems and data each one touches
- An audit trail for every agent that replaces it
FAQ
Questions we get asked.
Is this a security tool?+
Partly. Discovery gives your security team the visibility they've been missing: unsanctioned tools, personal accounts, and data going where it shouldn't. But the point is what comes next — turning the valuable work into agents the company owns, instead of shutting it down.
Do we have to install anything?+
Not to get started. The first discovery works from the activity your systems already record. If you want to go deeper once you've seen what's there, we'll talk it through with you first.
Are you reading everyone's AI conversations?+
No. Discovery starts from activity — which systems, which actions, how often — not from what people type. When we want to understand a piece of work properly, we ask the person doing it, and they know they're being asked.
Will you stop our people using AI?+
No. Blocking tends to push the work further out of sight. We find it, then help you bring it in properly, so people keep the time they were saving.
What happens to the work you find?+
You decide. For the work worth keeping, we rebuild it as a tested, secure agent that runs on company credentials and belongs to the company, not to whoever set it up. The rest you can review, approve or retire.
How do we start?+
Book a demo. We'll show you what discovery turns up and what it needs from your side, then agree a first step that fits.
Book a demo
See the AI work already happening in your company.
Tell us a little about you and we'll be in touch to set up a time.
- 01See what a discovery report shows
- 02Talk through your stack and where the risk sits
- 03Leave with a clear first step
Tell us about you
under a minute