Part of our ongoing series evaluating AI tools through a B2B lens.
Most AI tools stop at the edge of the screen. They draft the email, suggest the next step, and leave the actual phone call to you. Asmi is built to cross that line. It makes real outbound calls on your behalf: to the dentist, the bank, the plumber, the insurer, the contractor who went quiet. It navigates the phone tree and the hold music, then texts you the result on WhatsApp or iMessage.
The interaction model is unusual. Every morning, Asmi calls you. You talk, dump whatever’s on your plate, and it works the phones through the day. It launched publicly in mid-2026 out of Humint Labs. The founding team pairs CEO Rishi Raj Rathore, previously a co-founder of Arzooo (a B2B retail-tech platform he helped scale to roughly $400M in revenue), with Satwik Kottur, a CMU PhD and former Meta AI research scientist who also spent time at DeepMind. An earlier version billed itself as an “AI Chief of Staff” for inbox and calendar; the current wedge is narrower and sharper: the calls nobody wants to make.
One frame to set up front. Asmi is a tool for one busy person, not a team. Read it that way and it holds up.
What It Does Well
- Execution over the phone, the hard part. Anyone can draft a script. The work is the hold music and the back-and-forth with a receptionist, done without bailing at the first branch. That’s where most “agent” demos fall apart, and it’s what Asmi is built for. Early users report it handles those messy stretches.
- Voice-native, near-zero setup. No app to install, no dashboard to learn. You talk and it acts, then a text lands when the job’s done. For a non-technical user, the learning curve is close to flat.
- Memory and follow-through. Asmi retains tasks across days and weeks. Tell it Monday you need a dentist appointment; if the office calls back Tuesday with slots, it can handle that callback on its own. That persistence is what separates it from a reminder app.
- Multilingual reach. It reportedly speaks 50-plus languages, which matters for anyone coordinating across regions or vendors.
How Each Role Puts It to Work
- Business Leader / founder: the intended user, and the strongest fit. This is the person Asmi was built for. Hand it your morning brain-dump and it clears the personal-admin queue that eats a founder’s week: chase the accountant, dispute a charge, follow up with three vendors who ghosted, call five resorts to price a team offsite. The Monday-morning version is plain. Hand it the call list while you run the company.
- Seller: real but bounded. A rep drowning in logistics can offload coordination calls, like confirming a venue for a client dinner or chasing a no-API supplier for availability. It is not a prospecting or discovery tool, and you shouldn’t point it there. An AI voice cold-calling buyers creates brand and compliance headaches. It’s an admin helper here, useful for the logistics around a deal but not the selling itself.
- Finance: narrow, with a hard ceiling. It could call a vendor to chase a payment status or dispute an invoice line. But finance runs on auditability and control, and there’s no evidence of the access logs or approvals that implies. Letting an AI handle account-level calls that involve identity checks or payment details is a risk you’d want to keep on a human’s desk.
- HR / People: mostly not the tool. Logistics only: book the offsite venue, confirm the catering. Anything touching candidates or sensitive employee information should stay human-led, and Asmi offers none of the role-based access or record-keeping an HR team would need.
- Marketer: weakest fit. It won’t touch content or campaign work. At most it calls venues and vendors for event logistics. It isn’t a marketing tool, and calling it one would be padding.
Where It Could Be Better
The gaps are mostly a function of stage. Pricing isn’t public, which makes evaluation hard before you commit and makes finance sign-off harder still. Some businesses will refuse to deal with an AI caller, or will require a human authorization step. High-stakes calls (medical, legal, anything emotionally loaded) should stay human, and Asmi is honest that it’s aimed at low-creativity, high-friction chores. It’s early, too: a small team, a thin independent track record, and no admin or integration layer to speak of.
Why Not Just Use ChatGPT, Gemini, or Claude?
Because none of them will pick up the phone. That’s the entire point, and it’s a legitimate one. The founder’s own origin story is having Claude, ChatGPT, and Gemini open in three tabs while a task sat undone because it needed a real call. General models can reason about the to-do list all day, but not one of them will dial. So Asmi does something they can’t.
The caveat is what’s underneath. Asmi runs on Gemini, Twilio, and ElevenLabs, so there’s no proprietary model here. The value is the orchestration, the memory, and the telephony plumbing that wires those pieces into a working call. You’re paying for the workflow, not the model. But it’s a workflow you can’t assemble from a chatbot subscription alone.
Security & Compliance
Humint Labs is the named operator, and its terms address the right instincts: AI disclosure when calling on your behalf, consent and recording rules, blocked call categories, and call-timing limits. That’s more thoughtful than most consumer tools. What’s missing, as far as anything public shows, is the enterprise substrate. No evidence of SOC 2 or ISO 27001, no stated GDPR or HIPAA posture, no SSO/SCIM, role-based access, or audit logs. Absence of evidence isn’t proof of absence. But a finance or IT approver will treat it as an open question and won’t find a box to check. Note too that regulations like the EU AI Act require voice agents to identify as AI at the start of a call, which puts the disclosure obligation on the deployer. Treat sensitive calls (identity checks, payments, account changes) as human territory until those controls are confirmed.
Data & AI Connectivity
Asmi’s clever move is that it needs no integrations to do its core job. It reaches the phone network the way a person does, so a business without an API is still reachable. Its channels are voice, WhatsApp, and iMessage, with the earlier product touching inbox and calendar. What it doesn’t do is plug into your stack. No public API or webhooks, and no CRM or warehouse connectors turned up in research. It stays sealed off from your own systems even while it can reach any phone line in the world. That works for a personal assistant, but it’s a limit the moment you want to wire it into a team workflow.
Ratings
| Dimension | Rating | Rationale |
|---|---|---|
| Usability | 4.0 / 5 | Talk and it acts, with almost no learning curve; docked for early-access friction and unclear onboarding. |
| Power | 3.5 / 5 | Handles the hard part (live phone execution) well, but narrow and unproven at scale. |
| Flexibility | 2.5 / 5 | Serves one role and one channel; no team, admin, or integration breadth. |
| Cost | 2.5 / 5 | No public pricing to evaluate, which is real friction for smaller teams and for finance sign-off. |
Best-Fit Roles
Strongest for the individual founder or operator; useful to sellers for coordination admin only; marginal for finance, and off-target for HR and marketing as team tools.
Conclusion
Asmi is doing something most “AI agents” only claim to: acting in the physical world, across the one interface that still has no API, the phone. That’s a real wedge, and for a founder buried in call admin it can hand back hours a week. Just size it right. In our read, this is a personal delegate, and the trust scaffolding a business needs before it touches money or people isn’t visible yet. Point it at reversible, low-stakes chores, and keep a human on anything that involves an account or a real negotiation. Aimed at the right chores, it hands back hours. Point it at the calls that needed a person, and a polished AI voice just delivers the mistake faster.
Last updated: 7/8/2026.