Inside Two AI-Native Companies: An Accounting Firm and a Revenue Agent

We have argued elsewhere that the next decade of company-building belongs to teams that treat AI as the operating system they run on, not a tool they bolt onto existing workflows. That essay is here.

This is a different kind of post. Less theory, more receipts. Two companies, two industries, two completely different workflows — and one platform underneath. The Autonomous is the same product for both. The agents, the memory, the orchestration, the BYOM gateway, the WhatsApp and Telegram entry points — all shared. What changes is which agents are doing the work and which tools they reach for.

That is what an AI-native company looks like in practice. The company picks the shape. The platform makes it real.

Case Study 1 — JAA Associates: An AI-native chartered accountancy firm

Industry: Chartered accountancy · Team size: ~12 · Where they live: Bengaluru, India · Channels: Telegram, the admin dashboard

JAA Associates is a chartered accountancy firm serving small and mid-sized businesses across audit, taxation, and compliance. Like most CA firms in India, their week revolves around a relentless cadence of client deliverables — GST filings, TDS returns, balance sheet finalisations, audit fieldwork — all of which depend on the firm's associates submitting accurate timesheets on time.

The problem they came to us with was deceptively boring. Every Friday, the partners spent hours chasing associates over WhatsApp: did you submit your timesheet? Associates would forget. Reminders would get lost in DMs. By Monday, billing was behind, and the partners were doing the same chase again. It was the kind of thing every CA firm in the country deals with — and the kind of thing no off-the-shelf tool solves cleanly, because the workflow is specific to how a CA firm bills, books, and reports.

What we shipped together

We deployed a Telegram-native reminder loop on top of The Autonomous, tied to JAA's roster, calendar, and admin dashboard. The flow looks like this:

• Every associate runs /link their.email@jaa-associates.com once. Their Telegram chat is now bound to their HR record.
• On a cron the firm configures from the admin UI (default: Daily 5:00 PM IST), the platform scans for outstanding timesheets for the current ISO week and pings only the people who have not submitted.
• Associates reply DONE to mark the period submitted, or HELP to flag a blocker.
• When someone sends HELP, the platform writes a priority notification into the partner's admin inbox and pings the firm's designated point of contact on WhatsApp. The partner sees, in one place, every associate who needs attention — without having to reconstruct it from three separate chats.

None of this is glamorous. That is the point. The compliance loop closes by itself. The partners get their Friday afternoons back. The associates get a single channel they actually check.

Why this is an AI-native company, not just automation

A naive automation tool could send a reminder on a schedule. What makes JAA an AI-native firm is what happens after the reminders:

• Every interaction — every HELP, every late submission, every blocker — is captured into the firm's shared memory. The CEO agent and the role agents read from it. Over a quarter, the firm starts to learn which associates need different cadence, which clients drive the most late submissions, which periods are structurally hardest.
• When the partner asks the platform a question in chat — "who is behind on this week?", "what was last quarter's compliance rate?" — the agent answers from that memory. No spreadsheet pull. No re-asking the team.
• The firm can extend the loop with more agents — a GST reconciliation agent for the April 2026 IMS rule, an AR-chase agent for client billing, an audit-prep agent for season — without rebuilding the substrate. Each new agent reads from the same brain.

The timesheet workflow is the wedge, not the ceiling. JAA can run their entire compliance practice on this surface as we ship more agents into it. That is what "AI-native" means in production: the company is shaped to learn, not just to execute.

Case Study 2 — getsoma.store: An AI-native revenue agent

Industry: E-commerce (Shopify) · Team size: Founder + small ops team · Channels: Admin dashboard, Shopify Admin API

getsoma.store is a direct-to-consumer Shopify merchant in a crowded category. Their problem was not chasing people. Their problem was velocity. A modern e-commerce store lives or dies on the catalog: titles, descriptions, tags, collections, prices, promotional copy. Each of those decisions is small. The number of them is not. A merchant who is on top of every product page is a merchant whose conversion rate moves; a merchant who is not, is a merchant whose competitors out-list them on the same search queries.

They came to us because catalog maintenance was eating their week. Editing 30 product pages in the Shopify admin was a half-day. Doing competitor research for a category launch was another half-day. Doing both of those every week, while running ads and fulfilment, was the bottleneck.

What we shipped together

We turned The Autonomous into a revenue agent for their store. The agent has two surfaces:

• Competitor insights. The merchant clicks one button. The agent surveys their category — competitor product pages, pricing, positioning, common claims, gaps — and returns a ranked list of suggestions with the prompt prefilled. Each card has the explicit edit it recommends: a tag to add, a description to rewrite, a missing keyword cluster, a competitor price band they are out of. Category-aware. Not generic.
• Prompt-driven catalog edits. The merchant describes the change in plain English — "Add the 'clean-ingredients' tag to every product in the Hair Care collection that doesn't already have it" — and the agent plans the exact mutations against the Shopify Admin API, shows them in a diff, and applies them only after the merchant approves. Each apply is sequential and rollback-safe. If something fails midway, the agent stops and tells the merchant exactly which products were modified.

The result is the same kind of compression as the JAA loop. What used to be a half-day of admin work is now a ten-minute review. What used to be impossible — competitive research at the depth of a category specialist, on demand — is now a button click.

Why this is a revenue agent, not just a Shopify editor

The natural framing for this product is "a faster Shopify admin." That framing under-sells what is actually happening. The agent is not a faster keyboard. It is the merchant's judgment, amplified.

• Every edit teaches the agent.Approved suggestions, rejected suggestions, and modified suggestions all flow back into the merchant's lessons graph. The next insights run is sharper because the merchant's brand voice and risk appetite are now part of the brain.
• The same agent extends to other revenue work. A merchant who runs an insights pass once a week starts asking the agent harder questions: which collections underperform on weekends?, which descriptions correlate with returns? The platform answers those from the data it has captured. No new tool. Same agent, smarter context.
• It is multi-channel by default. The merchant can kick off an insights run from the admin dashboard, or from Telegram, or — once the inbound surface is live — from a WhatsApp message between meetings. The agent does not care which channel the human is on.

Same platform. Different shapes. That is the point.

The two companies above could not look more different on the surface. One is a chartered accountancy firm in Bengaluru working in Telegram and Excel. The other is a Shopify merchant working in an admin dashboard and the Shopify Admin API. They have no shared buyer, no shared tooling, and no shared cadence.

They do, however, share an architecture. Both are running on the same multi-tenant platform. Both have a CEO orchestrator agent routing inbound messages. Both write every run into a tenant-scoped knowledge graph, and every agent reads from that graph before it executes. Both use Claude Sonnet 4.6 as the default brain, and either could swap to OpenAI or Gemini or a fine-tune without a migration. Both write lessons after every run so the next run is smarter than the last.

That is the bet. We do not believe the future of AI is one vertical-specific copilot per industry — we have seen how many failed AI sales agents and abandoned AI bookkeepers there already are. We also do not believe the future is a thousand isolated chatbots. The future is a small number of horizontal platforms that let any company assemble the workforce it needs, in the channels its people already use, with the model it already trusts.

Two companies in two industries proving that thesis is a start. The next ten will be different again. We expect that. The platform is designed for it.

If your company should be on this list

We are taking on design-partner customers in two phases. If you are a services firm — CA firm, law firm, audit firm, consultancy — with a workflow that lives in WhatsApp and spreadsheets, we want to talk. If you are a Shopify merchant or a D2C operator whose catalog is eating your week, we want to talk. If you are somewhere else entirely and you can describe one workflow that costs your team ten hours a week, we still want to talk — that is usually the right starting wedge.

Enter your website at theautonomous.org to see what your workforce would look like. Or email abhinav@chainflux.com with one paragraph on what you would have an agent do tomorrow if it could.

The era of the autonomous company is here. You pick the shape.