Software Engineer
The person who builds the agent is the person who sits with the client using it. You'll ship into production in your first month and carry the pager for it.
About Clientell AI
Clientell AI is a San Francisco company. We run Salesforce and revenue operations for enterprise clients, hands-on, and we build the custom AI agents that live inside those systems to handle the work our clients' teams don't have time for. The pod that designs, ships, and stands behind this for our largest accounts sits in HSR Layout, Bangalore.
We're building that team on purpose: one pod, the same accounts, working together instead of handing tickets to each other. Those accounts include a Fortune 100 pharmaceutical company, a European home improvement retail group, and a US commercial fleet safety manufacturer. We'll share more account detail as your interview progresses.
We're seventeen people, and about to be more. The next few hires shape what this company becomes, not just what it ships this quarter.
About this role
We build AI agents on top of Salesforce, HubSpot, and whatever other CRM or ERP stack a client runs. Agents that do the work, not agents that tell someone what to click. That means someone has to go sit with the client, understand the workflow as it actually happens including the spreadsheet they use instead of the CRM, and build the thing that fits their business rather than a generic version of it.
That's this role. You embed with an account, build fast, and own what you shipped until it works in production and keeps working. You're not writing a service and handing it to an ops team. There isn't one.
The second half of the job is the part that decides whether this company scales: every time you build the third version of something, you turn it into a piece of the platform so nobody builds the fourth by hand. We're hiring four of you, and the difference between four engineers doing bespoke client work and four engineers building a system that does client work is the whole thing.
What you'll actually do
- Build and ship AI agents that plug into Salesforce, HubSpot, and whatever else the client runs, and do real work inside them
- Sit with a client's team, watch how they actually work, and figure out what to build before you write code
- Own the full loop: prototype in front of the client, harden it, deploy it, and be the person who gets the call when it breaks
- Build the evals. If you can't measure whether the agent is right, you haven't shipped it, you've demoed it
- Instrument what you ship: traces, failure logs, a way to know it broke before the client tells you it broke
- Design what the agent is allowed to do, and what happens when it's wrong. Non-deterministic systems writing to a client's system of record is the actual hard problem here
- Turn the third bespoke build into a reusable piece of the platform, so the fourth client takes a week instead of a month
- Work directly with the Salesforce Developers on the account so the agent understands the org it lives in
- Turn a vague complaint into a working demo fast, then decide what's actually worth hardening
- Make judgment calls with incomplete information, and be able to explain them afterward
What makes you a fit
- You can actually build: Python or TypeScript, real API work, and enough command of LLMs and agent tooling that wiring an agent into a live system isn't foreign territory
- You've put something with an LLM in it in front of real users, and you know exactly what it got wrong. Not a weekend project, something that had users who complained
- You build evals as part of shipping, not after. You have an opinion about what makes an agent unreliable and it's based on something that failed on you
- You've owned something in production. You've been paged, you've debugged under time pressure with a customer waiting, and you know what you changed afterward so it couldn't happen the same way twice
- You can read Salesforce's REST and Bulk API docs and get a call working without hand-holding. You don't need to be a Salesforce developer, you do need to not be lost
- You'd rather ship something rough in front of a client today than something polished in front of nobody next month
- You're comfortable being dropped into a mess with no spec, and comfortable being on the call where the mess gets explained to you badly
- You can explain what you built to a client's VP without an interpreter
- You debug like it's personal
- Excellent spoken and written English, most of your calls are with US teams
- Realistically this band covers zero to five years, and we care far more about what you have shipped than when you graduated. The bottom of it is for someone straight out who has clearly built more than that sounds like, the top is for someone who has already owned something real in production. Tell us on the first call where you think you sit
Use AI like you mean it
This matters more here than almost anywhere you'll interview, because it's literally the product. "I use ChatGPT sometimes" won't cut it.
Bring an agent or an automation you built. Two minutes, screen shared. We'll ask how you knew it was working, what happened the first time it wasn't, and how you found out. If the honest answer is "a user told me," say that, it's a better answer than a story.
Even better if
- You know Salesforce as an admin, a developer, or in any application-level role, enough that you're not lost inside someone else's org
- You've run observability on a non-deterministic system in production: traces, failure logs, alerting that fired before a user complained
- You've built the unglamorous infrastructure: queues, retries, backoff, rate limiting against a third-party API that lies to you about its limits
- You have a decent eye for UI or UX and can ship a frontend when what you built needs one
- You've worked on evals or model quality seriously, and you know why a benchmark number and a working product are different things
- You don't dread client calls
- You've sold or pitched anything to anyone
What it's actually like here
Some weeks are heavier than others. You fix something live in the middle of a demo, a deploy slips late because that's when the client's change window opened, a build gets thrown out and redone because the first version was wrong about what the client actually needed. When that happens everyone shows up. The rest of the time we don't perform busyness. If you want work that stops moving at 6pm every day regardless of what's in production, this will annoy you.
You'll carry the pager for what you ship. On-call is shared across the team, not dumped on whoever joined last, and we agree the terms with you before you go on the rotation.
What you get in return: real ownership from week one, your work in front of enterprise clients within the first month, no layer between you and the person using what you built, and four of you deep enough in the same problem that code review is actually useful.
Compensation and logistics
₹8-16 LPA based on experience and interview performance. If you're clearly better than the top of that band, tell us and we'll have that conversation. ESOP consideration as an early member of the team. Comprehensive health insurance included. First comp review at six months, not twelve.
Full time, in office, HSR Layout, Bangalore. Not remote, our delivery model depends on people overhearing each other's client calls. US client travel is possible. We'll walk you through what that involves, including visas, before you join.
How we're hiring for this
Four seats here, and six Salesforce Developers alongside it, plus interns on both sides. Same pod, same accounts. The Salesforce Developers own the orgs your agents run inside, so you'll be in each other's work constantly. If the agent gives a wrong answer in month six, it's usually because of a data model decision in month one, and both of you will be in that conversation.
Interview process
Three steps, not ten. We don't drag interviews out and we don't leave you wondering — you'll know where you stand after each one.
- Thirty minutes with the hiring manager: role fit, your AI demo, your questions
- A two-hour build session: a real problem, your own machine, your own tools, AI included. We care about how you get to something working, and how you'd know if it stopped working
- A conversation with the founder, we share more account context, you meet the people you'd work with
We aim to close this in under two weeks from the first call. If we pass, you'll hear why.
Apply
Apply for Software Engineer
Resume and a way to reach you is all we need to start. The rest helps, but skip anything that doesn't apply.