Ask a human: the human benchmark as one application with the Benchmark Agent
The request form for a human benchmark is no longer a page. It shares one title bar, one command bar and one brief pane with the agent, and nothing you type is lost.
Every buyer knows the moment when the model has done its job and you still want a person to look. The number is defensible, the percentile bars are drawn, and yet the deal is odd enough that you want a human analyst to check it by hand before you take it into the room. Until now, asking for that human check meant leaving the application and filling in a form that behaved like a form. Today we shipped the fix. The door to a person is rebuilt on the same application kit as the Benchmark Agent, so the two windows share one title bar, one set of menus, one command bar, one navigation and one brief pane. You do not change tools to change who answers. We wrote about why a human still earns the ask in humans are still more accurate than AI, and this update is the plumbing that makes that ask feel like the rest of the product.
One window, seven rooms
The request is organised into seven rooms in three groups. Under The request you have your paper, the deal, the negotiation and anything else. Under What you get you have what you want back and a call. Under Send you have review and send. The grouping matches the order a procurement buyer actually thinks in. You describe the situation, you say what deliverable you expect, then you send. The command bar and navigation are the ones you already learned from the agent, which is the same discipline we applied when the Negotiation Agent became a Windows application rather than a page. If you know one surface, you know this one.
The ten questions, and the four figures that watch them
The heart of the request is ten questions rendered as numbered rows with chips. Chips are not decoration. They let you set the shape of an answer without typing a paragraph, which is what keeps a request short enough to actually finish. Above the rows, the standing line names the next move, so you are never guessing what the form wants next. Along the top, four figures count in real time: how many papers you have attached, how many of the ten questions you have answered, what comes back as a deliverable, and the promised day. Those four numbers are the buyer's honest status. When the answered count reads ten and the promised day is set, you are done, and you can see it at a glance rather than scrolling to check.
That last point is the one buyers will feel most. The brief pane on the right is not a preview with different formatting. It is the same brief the human analyst opens on their side. What you wrote is what they read, in the order you wrote it. This is the same idea behind the Intake Desk that keeps your brief and now actually sends it. The gap between your intent and the analyst's understanding is where hand benchmarks go slow, and the shared brief closes it.
Nothing typed is lost
A request that loses your work is worse than no request at all, because you learn not to trust it. So this one holds. Nothing you type is lost across a reload, across the tab strip, or across tomorrow. Close the laptop mid answer and the request is where you left it. When you are ready to begin a fresh one, Start a new request shelves the last one under Recent requests, so nothing is deleted to make room. That persistence is the quiet feature that makes the whole thing usable for real procurement work, where a benchmark request is rarely finished in one sitting.
Keys, and where this sits in the workflow
For buyers who live in the product, the request is keyboard first. Keys 1 to 7 change rooms. Ctrl O adds a paper. Ctrl Enter sends. You can move from an empty request to a sent one without reaching for the mouse. This fits the way our command surfaces already behave, and if you have used Vera as a command desk the muscle memory carries straight over. The point of putting the human door on the same kit is exactly this: one set of habits covers both the model and the person. You reach for a human when the deal warrants it, not when the tooling forces a detour.
Honest limits
Be clear about what this update is and is not. It is a better front door to a human benchmark. It is not the benchmark itself, and it does not speed up the analyst's actual work. A person still reads your paper, still reasons about your deal, and still writes the finding by hand, which is precisely why you asked for a human. The promised day is a target set against real analyst capacity, not an instant result. If your deal is a clean fit for the model, the agent will answer faster and you should let it. This desk covers the request, not the answer, and it draws on the same benchmark library of 1,483 vendors that the agent uses, so the human is working from the same evidence you already trust. Send it a vague brief and you will get a slower, less useful answer, because the shared brief pane shows the analyst exactly what you gave them and no more. The tool removes the friction of asking. It does not remove the value of asking well. When the deal is odd enough to warrant a human, this is now the shortest honest path to one at /human-benchmark.
Want to be updated when major licensing and pricing changes land? One analyst brief a week: the price rises, metric changes and audit campaigns that move software costs. Work email only.
Morten brings two decades of enterprise and software procurement, with stints across Oracle, IBM, SAP, and Salesforce shaping how he reads a deal. He has led sourcing through hundreds of renewals, from mid market order forms to nine figure global agreements, and learned that the buyers who win are the ones who walk in knowing the market. He built VendorBenchmark to make that pattern recognition repeatable.
