The back office is the next automation frontier — and cross-border sellers should be paying attention
Most cross-border operators I talk to are drowning in the same unglamorous work that enterprise IT teams complain about: password resets, tool provisioning, onboarding a new VA in Manila or a 3PL contact in Shenzhen, offboarding a contractor who ghosted after two weeks. We obsess over ad creative and listing optimization while our own internal operations leak hours every single day. That’s why a launch like Harmony — an AI-native ITSM platform from the team behind Epsagon — deserves more attention from sellers than it will probably get. It’s not a Shopify app or an Amazon PPC tool. It’s a signal about where back-office automation is heading, and whether your current stack of Slack pings and Google Sheets is about to look very dated.
What Harmony actually solves — and who it’s really built for
The pitch from Ran Ribenzaft is refreshingly specific: legacy ITSM tools like Freshservice and Jira Service Management were designed to track work, not do it. That distinction matters more than it sounds. For the last decade, “ticketing system” meant a portal where an employee submits a form, a ticket gets created, someone triages it, someone else resolves it, and everyone feels productive because the SLA dashboard is green. The actual work — resetting the password, granting the app access, provisioning the laptop — still happened manually.
Harmony flips that. Employees ask in Slack or Microsoft Teams. Agents verify identity, take the action inside systems like Okta, Entra ID, and Google Workspace, and close the loop. No portal, no form. The company claims 100+ production-ready agents out of the box, plus a builder for custom workflows, with ticketing, asset inventory, SaaS management, onboarding/offboarding, and AI governance all in one workspace.
The numbers cited are aggressive: customers typically reach 60%+ auto-resolution within the first couple of weeks, Cyera is at 89% of employee requests auto-resolved, and n8n reportedly runs IT for hundreds of employees with a single IT person. Those are the kinds of claims I’d want to pressure-test in a pilot, but they’re also the kinds of claims that make a bootstrapped cross-border seller sit up.
Why Amazon sellers should care more than Shopify ones
Here’s my hot take: if you’re a Shopify DTC brand with a lean team and a handful of SaaS subscriptions, Harmony is probably overkill today. You don’t have an IT department, you don’t have Okta, and your “onboarding” is a Notion doc and a Loom video.
But if you run an Amazon FBA business with a Seller Central account, a Helium 10 seat, a Klaviyo account, a 3PL integration, a returns portal, a review automation tool, and three VAs across two time zones — you have an identity and access management problem whether you call it that or not. Every time you onboard a new ops person, you’re manually provisioning access to five tools. Every time someone leaves, you’re hoping you remembered to revoke everything. That’s exactly the surface area Harmony is targeting, just framed in enterprise language.
Where the math breaks
The counter-argument, and it’s a good one, comes from Konstantin Tikhaev in the comments: mid-market teams already use Okta or Google Workspace self-service for password resets, chat triage was never the real bottleneck, and the actual pain is audit compliance, rogue SaaS seats nobody deprovisions, and messy approval hierarchies. He’s right that an agent auto-resolving access across HR or Finance tools won’t have the verified approval authority an auditor needs.
For cross-border sellers, translate that to: the moment you’re dealing with VAT filings, customs brokers, or a payment processor asking for KYC documentation, “the AI did it” is not an audit trail. It’s a liability. The question isn’t whether the agent can take the action — it’s whether the action is defensible six months later when someone asks who approved it.
The architectural detail that actually matters
Buried in the comments is the most important technical point in the entire thread. Menni Alashvili, Harmony’s VP of R&D, explains that the model never executes actions directly — it only matches requests to pre-approved workflows, and every object it references has to resolve against real IdP or MDM data or the request simply fails. Scoped credentials, eligibility rules, human approval on sensitive actions, and a full audit trail mean prompt injection can steer interpretation but has nowhere to steer it to.
That’s the design pattern every cross-border seller should be stealing, regardless of whether you ever buy Harmony. If you’re wiring OpenAI or Anthropic into your customer service, your returns processing, or your supplier communication, the question you should be asking your team is: does the model have the ability to take an action, or does it only have the ability to propose an action that a pre-approved workflow then executes? Those are very different risk profiles, and most of the AI tooling being sold to sellers right now doesn’t make the distinction.
What a “pre-approved workflow” looks like for a seller
Concretely: an AI agent reading a customer email that says “please refund order #12345” should not have the ability to hit the refund API. It should be able to match that request against a rule — order is within 30 days, customer hasn’t refunded twice this month, product is not flagged as high-fraud — and then hand off to a workflow that executes. If any of those conditions can’t be verified against real data in your Shopify admin or your Amazon Seller Central order feed, the request fails and escalates to a human.
That’s not a new idea. It’s how good ops teams have always worked. What’s new is that AI makes it cheap to build the first layer — the interpretation and matching — without giving up the second layer — the execution and audit.
What cross-border sellers can borrow from this launch
Three things, in order of how quickly you can act on them.
First, treat your internal tooling like a product. The reason Harmony resonates is that it treats IT requests as a workflow problem, not a people problem. Most cross-border sellers I know run their internal ops on a combination of Slack, email, and tribal knowledge. That works until it doesn’t — usually the week someone quits. Spend an afternoon listing every recurring internal request your team handles (access provisioning, refund approvals, supplier follow-ups, listing edits) and ask which ones have a deterministic rule underneath them. Those are your automation candidates.
Second, steal the Slack-first interface. The reason Harmony’s agents live in Slack and Teams instead of a portal is that nobody wants to open a portal. Your VAs and ops staff already live in Slack or WhatsApp or Telegram. If you’re building any internal automation, build it where people already are. A Zapier or Make flow triggered by a Slack slash command will get adopted ten times faster than a custom dashboard nobody opens.
Third, take the audit trail seriously before you need it. The Harmony team’s emphasis on scoped credentials and full audit trails is the part that will matter most in twelve months when you’re explaining to a payment processor or a marketplace compliance team how a decision got made. Start logging now. Even a simple Google Sheet with timestamp, requester, action, approver, and outcome is better than nothing.
Where I think Harmony itself falls short for sellers
Two honest caveats. The integrations listed — Okta, Entra ID, Google Workspace, and 100+ others — are enterprise IT systems. None of them are Shopify, Amazon, TikTok Shop, Temu, SHEIN, Etsy, or eBay. If your “IT requests” are actually marketplace ops requests — reinstating a suppressed listing, appealing a policy violation, chasing a lost FBA shipment — Harmony doesn’t solve that today. It solves the layer underneath, which is employee identity and access.
And the pricing isn’t disclosed on the launch page, which for a bootstrapped seller is usually a signal that it’s priced for companies with an IT budget line item. That’s fine. Harmony isn’t trying to sell to a two-person dropshipping operation. But it means the value for most readers of this blog is in the pattern, not the product.
What I’d watch / test next
This week, do three things. First, audit your own internal request volume — count how many Slack messages or emails last month were someone asking for access, a password reset, or a tool provisioning. If it’s more than a handful, you have a problem worth automating. Second, look at one AI tool you’re currently using (customer service, listing optimization, supplier comms) and ask the hard question: can this thing take an action, or does it only propose one? If it can take action, you need an audit trail yesterday. Third, if you’re on Okta or Google Workspace and running more than 20 people, book a demo with Harmony — not because you’ll necessarily buy, but because seeing how a security-first agentic workflow is architected will change how you evaluate every other AI tool you’re pitched this quarter. The back office is the next automation frontier. The sellers who figure that out first will be the ones who scale past the point where tribal knowledge stops working.
