Cignara Wants Its AI Agents to Answer the Phone - and Actually Get Things Done

Call a big consumer brand about a broken appliance or a bill you do not recognize, and you already know the choreography. A menu that does not have your option. Hold music. A bot that offers to look up your order status when what you want is a refund. Cignara, a San Francisco startup in Y Combinator's Spring 2026 batch, is built on a simple observation about that experience: the software you talk to almost never has the authority to actually help you.

The company builds AI voice and chat agents for large business-to-consumer enterprises - the kind of firms that field millions of support calls and messages a year. The pitch, printed plainly on its site, is "agentic AI that understands your business, not just FAQs." The distinction sounds like marketing until you look at what most support automation still does, which is match a question to a canned answer and, when that fails, drop the customer into a queue for a human.

Cignara's argument is that answering is the easy half. The hard half is acting - issuing the refund, applying the policy exception, updating the account, recognizing that the caller who wants to cancel might stay for a different plan. Its agents are designed to do those things end to end, with what the company describes as governed, verifiable, policy-bound actions grounded in a company's own knowledge base and workflows.

01 / THE PROBLEMDeflection is not the same as help

The support-bot industry has spent a decade optimizing for a metric called deflection: the share of tickets resolved without a human. It is a cost metric, and it has a perverse edge. A bot can hit its deflection target by handling every simple question and shoving everything hard onto a person - which means the customer with a real problem gets the machine first, then the wait, then the human. The automation absorbs the easy work and passes along the friction.

Cignara's founders frame the gap in blunt terms. Most chatbots, they argue, cannot understand a customer's profile, cannot follow a complicated enterprise policy, cannot execute a multi-step workflow, and cannot notice when a conversation is quietly turning into a chance to keep a customer or grow the account. Those four gaps are the whole product thesis.

There is a reason those four gaps have stayed open for so long. Answering a question is a language problem, and language models are good at language problems. Taking an action is an operations problem - it touches billing systems, entitlement rules, fraud checks, and a thicket of exceptions that live in a company's head rather than its help center. Bridging that gap is less about a cleverer model and more about the plumbing that connects a conversation to the systems where things actually change. That plumbing is unglamorous, which is probably why it has been the last part to get built.

02 / THE PRODUCTAn agent that acts, and a copilot that whispers

The system has two faces. In autonomous mode, the AI agent takes a phone or chat conversation from hello to resolution, executing the workflow and taking the action the situation calls for. In copilot mode, the same underlying model rides alongside a human representative in real time, surfacing the right answer, the relevant policy, and the next step so a new hire can respond with the speed and accuracy of a veteran. An enterprise can run either, or both, and shift the dial as its comfort grows.

What makes those two modes possible is the layer underneath, which Cignara describes as building an enterprise's "AI brain." Before the agent ever answers a call, the company ingests the knowledge base, the business logic, the process workflows, and customer profile data. The result is not a chatbot pointed at a help center - it is a model of how a specific company actually operates, which is what lets the agent act within policy rather than improvise.

That last step is the part that separates Cignara from the deflection crowd. Where most tools sell support as a cost to shrink, Cignara sells it as a place revenue can appear. A caller phoning to complain is also a caller you can retain; a routine question can surface the right add-on. Handled well, the support line stops being a drain and starts nudging customer lifetime value upward - the kind of framing a chief revenue officer, not just a support director, will sit up for.

03 / THE FOUNDERFrom self-driving shuttles to self-driving support

Cignara is the second Y Combinator company for its founder and CEO, Nalin Gupta. The first was Auro Robotics, a self-driving shuttle startup he co-founded roughly a decade earlier and ran through YC's Summer 2015 batch before it was acquired by the mobility platform Ridecell. In between, he collected a Forbes 30 Under 30 nod and an engineering degree from IIT Kharagpur.

The through-line from autonomous vehicles to autonomous support is not obvious until you say it out loud: both are bets on letting software take consequential actions without a person holding its hand. A shuttle that drives itself and an agent that resolves a billing dispute share the same design problem - how do you make a system trustworthy enough that a large, cautious organization will let it act. That is a founder who has spent his career on the trust side of automation, not just the capability side, and it shows in how Cignara describes itself: governed, verifiable, policy-bound. Those are not words a demo-obsessed team reaches for first.

There is one more tell in the paperwork. The company's LinkedIn page still lives at the handle for "Bujo AI," the name Cignara carried before its 2026 rebrand. Renames in a startup's first year are easy to tease, but the more useful reading is that the product found its footing faster than the branding did - the thing being sold to LG did not change when the sign on the door did.

Repeat founders also tend to sequence a company differently. First-timers often chase capability and worry about distribution later; someone on their second attempt usually knows that the hard-won asset is a customer willing to vouch for you. Landing a recognizable enterprise before a public launch, rather than after a splashy funding round, reads like a founder optimizing for proof over noise. It is a quieter way to start, and in a category this loud, quiet can be its own kind of signal.

04 / THE MARKETA crowded, well-funded room

Cignara is not wandering into an empty field. AI customer support is one of the most contested categories in enterprise software right now, with heavily funded players - Sierra, Decagon, Cresta, Parloa, PolyAI and Ada among them - all pitching some version of agents that resolve rather than deflect. The incumbents are moving too, from Salesforce's Agentforce to Intercom's Fin. Being early is not the moat here; nearly everyone is early.

So where does Cignara try to stand apart. Two places. The first is the revenue framing - selling support as a growth surface, not only a cost center, which points the product at upsell, cross-sell and churn-prevention rather than pure ticket deflection. The second is the emphasis on governance for large B2C enterprises specifically, where a wrong action is not an awkward reply but a mispriced refund, a broken policy, or a compliance headache. For a Fortune 500 brand, the ceiling on adoption is rarely how clever the agent sounds; it is how confidently the legal and operations teams can predict what it will and will not do.

That a name like LG appears among its users is a meaningful signal at this stage. Large consumer brands do not casually put an early startup's AI in front of their own customers, and a logo of that size tends to say more about a young company than any funding announcement. It is the difference between a promising demo and something running where mistakes would be visible.

05 / WHAT YOU CAN DO WITH ITWho it is for, and why

Cignara's natural customer is any business-to-consumer operation drowning in conversation volume - telecoms, retailers, consumer electronics, financial services, anywhere a contact center is both a major cost line and a place customers form their opinion of the brand. For those teams, the offer is concrete: deflect fewer hard problems onto people, resolve more of them automatically, and turn the human staff you keep into a higher-performing group by giving each rep a copilot that knows the whole rulebook.

For a support leader, the appeal is coverage without a linear headcount curve - agents that handle the 2 a.m. spike and the routine 80 percent, freeing people for the genuinely thorny cases. For a revenue leader, it is the prospect of a support channel that occasionally pays for itself. And for a customer, at least in theory, it is the rarer thing: reaching something on the first try that can actually fix the problem instead of promising to transfer you.

The open questions are the ones every company in this category faces. How well does an "AI brain" hold up when a company's policies contradict each other, or when a customer says something the workflow never anticipated. How much human oversight does governed, verifiable action really require before an enterprise trusts it unattended. And how does a small team in San Francisco keep its edge as the incumbents bolt similar features onto software their customers already run. Cignara has not published the answers, and at this stage no one in the field convincingly has.

The company is still small and early - a founding team in San Francisco, actively hiring its first engineers, with the usual unknowns of a business this young. What is not unknown is the shape of the bet. Cignara is wagering that the next phase of customer support is not better answers but real actions, taken by software an enterprise can trust, and that the founder who learned to make machines act on public roads is a reasonable person to make that wager twice.