Ressl AI Wants to Run the Office Nobody Wants to Run
The San Francisco startup, part of Y Combinator's W26 batch, deploys AI agents that handle the copy-paste admin behind trades and home-services businesses. No new software. No migration. Just the boring work, done.
Walk into almost any plumbing, HVAC or collision-repair shop and you will eventually meet the person who holds the whole thing together with a keyboard. They are not turning wrenches. They are copying a phone number from a voicemail into a scheduling app, pasting an address into an estimate, then re-typing that same estimate into an invoice. Ressl AI has a name for this person: the glue employee. The company's whole bet is that software, not another hire, should do that job.
Ressl AI is a San Francisco startup in Y Combinator's Winter 2026 batch. Its founders, Arushi Gandhi and Abhishek Eswaran, describe the product plainly: AI agents that sit on top of the software a business already uses and handle the admin autonomously. Answer the lead. Draft the estimate. Chase the part. Reconcile the invoice. The pitch is less about intelligence for its own sake and more about the unglamorous middle of a business, the part that rarely shows up in a demo reel but quietly decides whether a shop grows or drowns.
The problem is easy to underrate until you have lived it. In field operations, a missed call is often a lost job. Estimates that take two days lose to the competitor who replied in twenty minutes. And the office that is supposed to keep all of it moving becomes a bottleneck the moment volume rises, because the work scales linearly with people. Hire more admins, or watch leads fall through. That is the trap Ressl AI is trying to spring.
01 · What it actually doesThe office, on autopilot
The core idea is that agents should live where the work already happens. Ressl AI does not ask a shop to rip out its scheduling tool or migrate its records into a new system. It layers on top. When a lead comes in over a call, a text, a Yelp message or a website form, the agent responds across those same channels, books the job, and then moves down the chain: generating an estimate, coordinating with parts vendors, and closing the loop with invoicing and follow-ups. The human stays in the loop for judgment. The machine takes the repetition.
It is worth saying that Ressl AI did not start here. Its earliest agents worked inside Salesforce, letting IT and RevOps teams configure the platform in plain English - documenting an org, root-causing a support ticket, mapping dependencies, even writing Apex code and test classes. That work taught the founders something useful about deploying agents into systems where mistakes cost money. The trades pivot took that lesson somewhere with less software sophistication and far more pain.
02 · Who it's forThe customer wears work boots
Ressl AI's users are HVAC, plumbing, electrical, landscaping, roofing and collision-repair businesses - the companies that keep the physical world running and rarely make a technology headline. There is a second, quieter customer, too: the private-equity firms buying up home-services companies and rolling them into larger operations. For a PE-backed roll-up, a back office that runs on agents rather than headcount is not a convenience. It is the margin.
The founders are an unlikely-on-paper pair that makes sense in practice. Arushi Gandhi, the CEO, is ex-Microsoft and a Generation Google Scholar who, before Ressl AI, built an AI CRM to roughly $120K in annual recurring revenue. Abhishek Eswaran is an IIT Bombay engineer who has sold a machine-learning model and scaled an AI services firm. One has shipped a small commercial product end to end; the other has built the guts. That combination shows up in how the company talks - equal parts go-to-market and architecture.
03 · Why it's differentLayer on top, don't replace
Plenty of companies now say the words "AI agent." Fewer sell to a plumbing-shop owner who has no interest in a platform migration and every interest in answering the phone faster. Ressl AI's differentiation is less a single clever feature and more a set of stubborn choices: run on top of existing software, require no migration, and get measured on work completed rather than seats filled. For a non-technical buyer, that removes the two things that usually kill an enterprise sale - switching cost and risk.
The traditional alternative to all of this is simple and expensive: hire another person for the front office. Ressl AI is effectively arguing that a lot of org charts are about to get shorter, not because people are being replaced wholesale, but because the copy-paste layer between people and software can now be handled by software. Against horizontal "AI employee" platforms and vertical field-services incumbents adding AI features, Ressl AI's edge is focus - one messy workflow, done reliably, for buyers the glossy startups tend to skip.
04 · The hard partMaking agents you can leave alone
Here is the honest difficulty in this category: agents demo beautifully and break in production. An agent that mis-schedules a job or fumbles an invoice does not just annoy a user; it costs a small business real money and real trust. Ressl AI's more recent positioning leans into this directly, describing a platform to train, evaluate and deploy autonomous agents - moving them from prototype to something that survives contact with a live business. The order matters. You earn the right to run unattended by proving reliability first.
That framing is a useful tell about where the company thinks the moat is. Not the flashiest agent, but the most trustworthy one - the kind a shop owner can point at money-touching work and then stop watching. Evaluation is unglamorous engineering. It is also, in a market full of impressive demos, the thing that decides who keeps a customer past month two.
05 · How it makes moneyPriced against a hire, not a seat
The business model follows from the pitch. Ressl AI is not really selling a login that a shop pays for per user each month; it is selling completed work, positioned as the alternative to adding a person to the front office. That distinction changes the sales conversation. A per-seat tool competes with other software budgets. An agent that answers leads and drafts estimates competes with a salary line, and a salary line is a far larger number. When the comparison is "another admin at $45,000 a year" versus "software that never sleeps and does not call in sick," the math tends to argue for itself.
It also explains the obsession with running on top of existing tools. Every migration a vendor demands is a reason for a busy owner to say "not now." By refusing to touch the underlying system of record, Ressl AI shrinks the decision to a single question: can the agent do the work? For the private-equity buyers assembling home-services roll-ups, that question compounds. A back office that scales with agents rather than headcount is the difference between a business that gets more expensive as it grows and one that gets more profitable.
The expertise underneath is the part that is easy to miss from the marketing. Building agents that touch money - scheduling real jobs, ordering real parts, sending real invoices - is a different discipline than building a chatbot. It requires knowing how systems fail, how to catch an agent before it makes an expensive mistake, and how to measure reliability rather than assert it. The founders' earlier work inside Salesforce, where a bad automation can quietly corrupt a company's records, was a hard school for exactly that. The trades are less technical as a market, but the tolerance for error is just as low.
06 · Where it fitsThe unsexy frontier
The center of gravity in AI right now is coding assistants and chat. Ressl AI is pointed at the opposite corner of the map: field operations, small businesses, back-office coordination. It is less crowded there, and the pain is more concrete. A shop owner does not need to be convinced that answering leads faster is valuable; they feel it every time a call goes to voicemail. If the bet is right, the biggest agent market may not be the one that looks most futuristic. It may be the one that smells like a garage.
Ressl AI is early - a handful of people, fast iteration, revenue that third-party trackers describe in modest, approximate terms. What it has is a clear customer, a clear enemy in the glue-work bottleneck, and a discipline about reliability that this category badly needs. Whether it becomes the default office for the trades or a well-run niche will come down to the least exciting question in software: do the agents keep working when nobody is looking.