What makes CIOs trust an AI agent? Thira bets it’s not the model.

Featured image credit: Tri wiranto for Unsplash+.

Sunny Gupta spent a decade and a half building Apptio into the system of record for enterprise technology spend. His next company is a bet that CIOs now want to act on that spend, not just measure it — and that the models are now up to the job.

Gupta and Apptio co-founder Kurt Shintaffer are building Thira, an agentic “system of execution” for the enterprise back-office, starting with IT. Ten enterprises have already signed on as unpaid design partners ahead of a launch this fall. The company is coming out of stealth with a $21 million seed led by Madrona.

Apptio provided the map

Like so often, the core idea behind Thira is based on complaints its founders heard during their last run. CIOs were grateful for the visibility, Gupta tells The New Stack, but still had todo actual work by hand. 

Gupta says he heard this on repeat in his later years of running Apptio, the technology business management company he took public in 2016 and sold to Vista Equity Partners in 2019 before it was acquired by IBM for $4.6 billion in 2023. 

Apptio, he says, “gave us the map.” 

What Thira does

With this map in hand, Thira now targets the back-office processes a CIO owns, starting with enterprise IT. Apptio sold to CIOs, so it only makes sense for Gupta to target them with Thira, too. 

The company remains tight-lipped about what its actually product will look like, but Gupta says the team mapped six or seven repetitive IT processes and will launch with just two, to “be the best in the world” at them before expanding.

The first process Thira is targeting touches more than 40 disparate systems on average, Gupta says, work handled today by people or ad hoc scripts. The plan is to launch the product by focusing on two of these processes and expand from there.

And while Thira isn’t saying much about the details of its product, Gupta says that a discovery and self-learning system is at the core of its platform. Thira’s agents connect to a company’s data sources, ingest documents, slide decks, and even hand-drawn diagrams, and learn how that particular environment is wired. From there they build what Gupta calls dynamic “execution maps.” Humans review and approve those maps, which set the deterministic paths agents can run on their own and flag the non-deterministic ones that still need judgment.

The trust problem Thira has to solve

Trust, Gupta says, has to be at the core of what the company is selling and in this agentic era, that means features like kill switches, audit trails, and rollbacks. “You can’t afford to have hallucinations when you’re touching IT systems,” he says.

Thira will offer two modes to its users. “A fully autonomous mode, once the enterprises get comfortable,” he says. “But also, to start with, if enterprises are not comfortable, we’re going to start them with a semi-autonomous mode, where a human can come in and still credential things, but we will log everything.”

Gupta believes it will take customers about a month to get comfortable with the semi-autonomous system before they will make the switch to full autonomy.

Not replacing ServiceNow

Thira isn’t out to replace the incumbents, Gupta notes. 

“ServiceNow is a great product. Workday is a really good product,” he says. “These are established systems of record. So we’re picking cross-functional processes which these companies are not really focused on.” 

The plan is to integrate with and extend those systems. ServiceNow CEO Bill McDermott, for example, wants his platform to be the “uber orchestrator,” Gupta notes, and Thira’s purpose-built agents can plug into that orchestration rather than fight it.

What that will look like in practice remains to be seen. ServiceNow has, after all, spent the past year opening what it calls its “system of action” to outside agents, and every enterprise vendor is seemingly chasing this same execution layer.

The hard part

The hardest technical problem in building Thira, Gupta says, is the learning engine and teaching agents to operate across the messy, differently configured systems inside each enterprise, where the same thing goes by a different name in each one. He’s betting on a flywheel he knows from Apptio, he explains: each customer benefits from what the system learns elsewhere, and the model improves once it has run across 40 or 50 of them.

Thira is also trying to run itself the way it wants its software to run. Gupta says nearly every function, down to the company’s website, will operate agentically, and that he plans to hire only two kinds of people: engineers now, salespeople later. The engineering team is seven or eight people, he says, and the whole company is just over 10. Thira’s team includes former engineering and product leaders from Atlassian, Meta, Databricks, and Oracle.

With a $21 million seed round, the team could expand quickly, but the plan is to stay lean. Madrona led the company’s funding round, making this Gupta’s third company Madrona Managing Director Matt McIlwain has backed (after iConclude and Apptio). FUSE and a group of prominent global advisors and CIOs also participated in this round.

Before joining The New Stack as its senior editor for AI, Frederic was the enterprise editor at TechCrunch, where he covered everything from the rise of the cloud and the earliest days of Kubernetes to the advent of quantum computing….

Read more from Frederic Lardinois