Computer-use agents and business software: why click-level workflow records are scarce

What is a computer-use agent, and what does it need to learn?

A computer-use agent is an AI system that carries out tasks in software rather than just answering questions about it. It reads a screen or an interface, decides on an action, performs it, checks the result and continues. Examples include filing a ticket, reconciling an invoice, updating a CRM record or routing an approval.

To get good at that, an agent needs to see how real work is done: the sequence of steps, the choices made, the exceptions and the outcome. That is a different kind of material from articles, forum posts or code snippets.

Why public data is not enough

Many large language models have been trained largely on public text. Researchers at Epoch AI have estimated the effective stock of public human-written text at roughly 300 trillion tokens and projected that, if trends continue, language models will fully use that stock sometime between 2026 and 2032. It is a forecast with wide uncertainty, and it concerns text volume, not workflow records, but it helps explain why developers look for non-public sources of training material.

Workflow records are thinner still. People rarely publish a complete, annotated account of how they approved a purchase order in an ERP or how a helpdesk escalated an outage. Such material is created in the course of work, stored inside company systems and protected by confidentiality terms.

What counts as click-level workflow data?

In plain terms, workflow data is a record of work as it happened, with context and results. For a company that means:

• A ticket with its full history: request, triage, reassignment, internal notes, resolution and customer confirmation.
• A purchase or sales order with approvals, changes and the final posting.
• A change request that links an issue, a code review and a deployment.
• An onboarding task list with who did what and when.
• A decision record with the alternatives considered and the outcome.

The term click-level describes the granularity an interface agent works at. In practice, most companies hold structured records and event histories, not screen recordings. What exists depends on each system. Do not assume a given tool captures every click.

Where those records live

Strong companies hold records across many systems, often 10-15+, and long histories of 5-10+ years or archived systems add depth. Connection between systems matters: a ticket linked to a change request and a deployment shows more than any one of them alone.

Why this matters to an IT consultant's clients

A mid-sized company's ERP, helpdesk and tracker histories are exactly the records that agent developers cannot get elsewhere. That does not mean every company's data has value: the history must exist, be exportable and be the company's to license, and the records must show real work rather than a thin trail.

The ITSM ticket history page covers one well-defined type. The warning in the guide on ROT data cleanup matters too: marking history obsolete and deleting it can remove the records an agent developer would value most.

What this means for a referral partner

You do not need to understand agent research to make a useful introduction. You need to recognize companies that keep deep, connected operational records and whose owners can decide to license them.

1. Spot candidates among clients with 50+ full-time employees at peak (contractors excluded), several years of documented operations and many connected systems.
2. Check the rights: the company must have created the records and be able to license them. See the SaaS export rights guide for the contract questions to raise with counsel.
3. Introduce the owner or an authorized sponsor through the referral form or your referral link.
4. Step back. SourceX qualifies the company, runs the inventory, agrees price and terms, manages buyer review and handles delivery. Partners never export, upload or describe confidential records.

SourceX is the enterprise data transaction layer for AI: it manages data licensing between companies and AI developers. SourceX does not train AI models. Companies keep ownership; data is licensed, not sold, and nothing is binding until the company signs. For a plain-language definition, see what is workflow data.

Limits and open questions

• Demand is not a promise. An agent developer's interest in a type of record does not mean any given company's records will be licensed. Qualification, rights review and buyer interest decide.
• Records vary in quality. Thin tickets, copy-pasted notes and templated entries carry little signal. Records generated with AI in order to sell them are a red flag.
• Privacy and confidentiality matter. Records can hold personal data or third-party information. De-identification and redaction requirements are agreed with the company before any work begins.
• Exclusivity. Deals are typically exclusive for AI training for an agreed term, which the owner must be willing to consider.
• The research is about public text. The Epoch AI estimate concerns the stock of public human-written text, and projections are uncertain.

Questions to ask a client before you introduce them

1. Which systems hold the day-to-day work, and how many years does each go back?
2. Do tickets, orders and approvals carry outcomes, or only a creation date?
3. Are the tools linked, for example a ticket to a change request to a deployment?
4. Who can run an export, and has anyone tested one recently?
5. Does the company own the records, or do they mostly describe its clients' work?
6. Is any tool about to be retired, so an export should be preserved first?

The answers tell you whether the 4-step introduction above is worth starting.

How rewards work

Partners earn 25% of the eligible platform fees SourceX actually collects from the referred company's licensing deals, capped at $100,000 cumulative per referred company. The reward is payable only after the buyer pays and SourceX receives its fee; a lead, meeting or signed agreement alone does not trigger payment, and no reward is guaranteed. It is never deducted from what the company receives. The role is a referral one, as explained in channel partner vs referral partner.

Next step

Think of one client whose helpdesk, ERP or tracker holds years of connected history and whose owner you can reach. Register as a partner and introduce them, or use the network opportunity finder to map several. MSPs can start from the page on referral opportunities for managed service providers and additional revenue streams for MSPs. Check the baseline on the who qualifies page.