Every company has tasks that repeat hundreds of times a month. Someone matches an invoice against the bank statement, answers the same question about delivery times, types up the notes from a field visit. For years, the way to keep up was to hire more people’s hours or buy more software licenses for those people to use.
AI agents changed that math. An agent can read the data, apply the rule, fill in the system and flag when something does not add up. When the whole task fits into that flow, it makes sense to buy the completed task instead of the hours or the access to a tool. That is what the market has started calling Task as a Service.
From hours to tasks
What companies buy in software has gone through four formats. They coexist in the same company, and each one answers a different question about who does the work.
| Model | What you buy | How you pay |
|---|---|---|
| People’s hours | Professionals’ time | Per hour or per person |
| Software as a Service | Access to a tool | Per user, per month |
| Task as a Service | Each task completed by agents | Per task |
| Service as a Software | The whole service, delivered ready | Per deliverable, outcome or subscription |
In 2024, Foundation Capital described this shift as the move from software as a service to service as software: instead of selling the tool and leaving the outcome to the customer, the vendor takes responsibility for the outcome. By their account, software starts competing for headcount and outsourced services budgets, not just the technology budget.
Some vendors already charge this way. Intercom, for example, charges for its Fin customer service agent per outcome, starting at US$0.99 each, according to its pricing page as of October 2026.
A task or a complete service
Task as a Service and Service as a Software go together. A complete deliverable, such as the monthly financial analysis for a business owner, is made of many tasks: collecting statements and invoices, reconciling each transaction, flagging discrepancies, deciding the cases outside the rules, building indicators and talking it through with the business owner.
Some of these tasks an agent can do on its own. Others need a person. Buying each completed task is Task as a Service. Buying the finished analysis, with all the tasks behind it, is Service as a Software. In practice, most companies start with the task, because it is easier to measure, adjust and trust one part before handing over the whole.
The traditional model and the agent model, side by side
Take the financial reconciliation of a mid-sized company. The difference shows up in who does the work, when it happens and what gets logged.
| Traditional model | With agents | |
|---|---|---|
| Who does it | A person, one transaction at a time | Agents that read statements and invoices |
| When it happens | At month-end close, in a batch | Throughout the month, as the data comes in |
| Where the person steps in | In all of the work | On discrepancies and on approving the close |
| What gets logged | Spreadsheets and email threads | Source, rule applied and who approved it |
| How you pay | Hours or licenses | Per reconciled transaction |
The person in finance stays in the process. Instead of typing and checking item by item, they decide what the agent is not allowed to decide: the invoice that does not match, the transaction with no document, the close that goes to the accountant.
Every task run by agents has a named person who approves the exceptions and is accountable for the result.
Where the person steps in
The most common concern for anyone running an operation is losing control. That is why the design starts with the limits, and only then with the automation. Before the agent runs, what it can read, what it can change and what it can send outside the company is defined in writing.
- Cases outside the rules stop and wait for a person to decide. The agent flags them, shows the data and suggests an answer, but does not close them on its own.
- Actions that cannot be undone, such as paying, deleting or sending to the end customer, require explicit approval.
- Every task leaves a record with the source data, the rule applied, what the agent did and who approved it.
It is the same approach Tog Lab uses in AI-assisted software development: the agent works inside a structure with context, rules and checks, and a person approves at the decision points. In operations, that structure is what lets you deliver a result to your own customers that you can stand behind.
How to choose the first task
Not every task is a good place to start. The ones that tend to work best share five traits.
- It repeats often, every day or every week.
- The rule is clear enough to be written down.
- The data it needs already lives in a system you can access.
- A mistake shows up quickly and can be fixed before it causes harm.
- Someone on the team has the authority to decide the exceptions.
Once the task is chosen, measure before you change anything: how long it takes today, how often it happens and how often it goes wrong. Those numbers will show whether the agent is delivering, and they help set the price per task.
How Tog Lab works with Task as a Service
The work starts by understanding the task with the people who do it every day. That produces a Domain Context with the rules, the exceptions and the risks. The agents are built on top of the systems the company already has, with the AI model it chooses, and the per-task pricing model is designed together with the client.
Once the first task is stable, the next ones join the same flow. Over time, they add up to a complete service that the company can offer its own customers as Service as a Software.
Compare the traditional model with the agent model
Sources
- Joanne Chen and Jaya Gupta, “AI leads a service as software paradigm shift”, Foundation Capital, Apr 19, 2024. foundationcapital.com
- Intercom pricing page, accessed Oct 27, 2026. intercom.com/pricing