Service as Software: what it is and where the idea came from

The first reading comes from venture capital. For Foundation Capital, AI companies are leading “a transition from software as a service to service as software,” and in the services business “responsibility for achieving the desired outcome sits with the company selling the service” (source 7). In 2026, Julien Bek of Sequoia Capital summed up the difference: a copilot sells the tool, an autopilot sells the work (source 10). In Sigil Labs’ reading, both are investor theses about where new companies will come from.

The second reading comes from the services industry. HFS Research spells the term “Services-as-Software” and, in a November 2024 article on services and software blurring together, describes “one blended, scalable solution we are calling Services-as-Software” (source 11). In Sigil Labs’ reading, the ones changing here are the services firms themselves.

According to Sequoia, the place to start is work that is already outsourced: the company accepts that it can be done externally, a budget line for it already exists, and the buyer already purchases an outcome (source 10). The essay adds that the autopilot wins sooner where work follows rules, even complex ones, and, on consulting, that the work is mostly judgment, leaving open whether AI can split off a layer to automate. In Sigil Labs’ view, judgment work whose result is hard to check is where a copilot serves better.

Some actions also call for a person in the loop. Singapore’s agentic AI governance framework recommends checkpoints that require human approval for high-stakes or irreversible actions (source 14).

The buyer is no longer the professional who uses the tool but whoever needs the outcome (source 10). Pricing follows that shift. In 2024, Sierra described pricing in which it gets paid only when it completes a task, with criteria agreed upfront for each outcome (source 12). In 2026, the same company wrote that this pricing only works where the software is highly autonomous and the outcome can be attributed to it, and that it is more complex than seat-based or consumption pricing (source 13).

In Sigil Labs’ view, for the client the question stops being how many licenses to buy and becomes what counts as work delivered, and who checks it.

Sigil Labs builds Service as Software systems from four parts: the process mapped with the client’s team, human approval points on the decisions that matter, limits for each agent, and an auditable record of what was done, by which agent and on what basis. Singapore’s framework points the same way: whoever deploys the agents remains accountable for their actions (source 14).

VoxxFlow, a Sigil Labs product in production, is one example: AI agents plan, create, review and publish marketing content, and nothing is published without human approval.

Empenho, a proof of concept, is the other: a platform for Portuguese accounting offices. Its supervised digital colleague leaves proposals in a “To review” queue and never sends anything, and the platform proposes extras to bill when it detects work outside the retainer, the contract with a monthly value, for someone in charge to approve.

Text written by the AI reaches the client only after someone in charge at the office approves it, and a number guard rejects notes containing figures that are not in the data given to the model. Beside the digital colleague, a copilot answers with sources, which puts both models on one platform.