Most of what gets called “AI automation” in business software is a chatbot that drafts an email and stops. Useful, sometimes. Not actually automation of anything.
Automode does something narrower and, we think, more honest: it runs the specific, repeatable arc from an incomplete inquiry to a priced offer, and it stops exactly twice to ask a human if it’s allowed to keep going.
What the flow actually looks like
An inquiry arrives without enough detail to price anything: no dates, no headcount, no real scope. Stark reads it, drafts a polite follow-up asking for what’s missing, and stops. Gate one. Nothing has been sent to anyone yet. You read the draft, click approve, and it goes out.
The client replies. Stark reads the reply, checks it against your company’s actual offer history in the Vault (what similar work has cost before, what line items typically apply, what your margins usually look like), and drafts a complete, priced offer. It stops again. Gate two. Still nothing has been created. You read the draft, the reasoning behind the pricing, and click approve.
Only then does a real Offer record exist, linked back to the original Inquiry, sitting in your pipeline exactly where it would be if you’d drafted it yourself.
That’s the entire feature. Two human decisions. Everything in between, reading the inquiry, drafting the request, reading the reply, pricing the work, happens without you, but never past you.
A real run, not a demo script
We tested this on a genuine inquiry, not a scripted one: a follow-up event request from a law firm client, missing dates and a firm budget ceiling. Stark drafted the request. We approved it. We pasted in a realistic reply: a budget around €8,000, dates confirmed. Stark drafted the offer: “3-Day Team Offsite,” three line items, €6,755, correctly under the stated budget, not just close to it. We approved it. A real Offer existed seconds later, in the actual pipeline, not a sandbox.
The reasoning is what we paid attention to, not just the number. It wasn’t a generic markup applied to a generic template. It referenced what this specific kind of work has actually cost this specific company before, the same grounding that makes Stark’s answers different from a generic AI tool’s answers everywhere else in the product.
Why two gates, specifically, and not zero or four
Zero gates means a system that can commit you to a client-facing promise without you ever seeing it: a non-starter, not a feature we’d want even if visitors asked for it. Four or five gates starts to feel like the automation isn’t actually saving anyone anything; you’re back to doing all the work, just with extra clicking.
Two is the number where the human stays in control of every consequential moment, what goes out the door, and what gets priced, without being asked to babysit the parts that are genuinely mechanical: reading a reply carefully, cross-referencing it against pricing history, drafting clean language. That’s the actual design argument, not an arbitrary choice.
It doesn’t stop at the offer either
Once the client accepts and the Offer is approved, Automode’s reach continues into the Project. The same context, the line items, the scope, the history that led there, carries straight through, so the Project starts already populated instead of being rebuilt from scratch by whoever picks up delivery. The two gates cover the two moments that actually need a human decision: what goes out the door, and what gets priced. Everything else, including the handoff into delivery, keeps moving without asking anyone to manually re-create what the system already knows.
What it isn’t
It isn’t full autonomy, and we don’t want it to be. It isn’t going to chase a client who never replies. That’s still a human’s call to make. And it doesn’t get smarter by guessing; if a reply genuinely doesn’t have what’s needed to price the work confidently, the draft reflects that honestly rather than inventing a number to look decisive. The two clicks aren’t a formality bolted on for optics. They’re the actual point.