When an agent earns its price, and who holds the key
The announcements on 12 November circle one question: once a tool carries an AI agent, how much does it actually change, and who controls the parts that cost money. Buyers are moving past the demo and asking about outcomes, pricing, and whether the work they build can leave with them.
That is the useful lens for the day. An agent is only interesting if it changes a number you care about, and the arrangements around it — who pays the model bill, who owns the reusable work, who runs the company behind the product — decide how much leverage you keep. If you track these notes daily, this one follows on from the 11 November briefing.
Bringing your own key, and the cost that sits behind an agent
When a tool runs an AI feature for you, something is paying for each model call in the background. Usually the vendor buys that capacity in bulk and folds the cost into your subscription, which is simple but hides the unit economics from you. Bring your own key, or BYOK, flips that: you connect your own model account, and the usage lands on credits and rates you have already negotiated.
Linear made this concrete for its cloud coding agent. The company says you can now bring your own key and “use your existing provider credits and negotiated rates”, connecting through providers you already hold accounts with [1]. The trade-off is plain. You take on managing the model relationship and watching the spend, and in return the cost stops being a flat line you cannot see inside. For a team already buying model capacity for other work, routing an agent through the same account avoids paying twice for the same tokens. For a team that wants one invoice and no metering to think about, the bundled approach is still the easier one. Neither is wrong; the point is that BYOK turns a hidden cost into a visible one you can shop around.
Whether an agent moves the contract value
Most AI features are added to a product without changing what anyone pays for it. They make an existing plan a little nicer. A smaller set of agents change the price itself, because they do work a customer would otherwise pay a person to do. That difference is worth being strict about when you evaluate a tool.
SaaStr described a portfolio company, a vertical software vendor at about “$30M ARR”, that launched its first real agent. The headline detail is blunt: “The agent doubles ACV”, and there is no free version of it [2]. ACV is average contract value — what a customer pays per year. An agent that doubles it is doing something a buyer treats as worth real budget, not a toggle thrown in to keep a renewal warm. The lesson for a business choosing tools is to separate the two. When a vendor shows you an agent, ask what it replaces and whether the vendor is confident enough to charge for it. A feature that cannot carry its own price is usually convenience; a feature priced on its own is usually doing work. This is the same discipline behind knowing which decisions automation should never make: an agent you pay for is one whose output you have decided to trust.
Customers are already sending agents ahead of themselves
The other side of the agent story is the one arriving at your front door. Your customers are beginning to use AI to research before they ever reach you, which changes what your storefront and your content have to be legible to.
Salesforce published survey findings for the holiday season. It reports that “holiday shoppers are using AI agents for research and advice — almost 20% of shoppers now say that AI is their first stop on the way to the store — but most aren’t ready to let AI click Buy Now” [3]. Read that carefully, because it points in two directions at once. A fifth of shoppers now begin with an agent, so the information an agent can read about your product — prices, availability, plain descriptions — is doing work before a human looks. At the same time, the purchase still happens with a person, so the handoff from agent research to human checkout is where the sale is won or lost. If your pages are easy for an assistant to summarise but hard for a person to buy from, you have optimised the wrong half. The same gap applies to lead handling; it is worth checking where your business loses leads between first contact and conversion.
Portable skills, and what stays behind
As teams build repeatable work for agents, a lock-in question appears. If the instructions you write only run on one vendor’s agent, you have quietly tied your process to that vendor. The industry is starting to separate the reusable part from the host-specific part.
GitHub’s guide to agent skills draws that line. It explains that “agent skills capture repeatable engineering work in files teams can review and reuse, while tools, project context, and permissions remain specific to the host that runs them” [4]. For a buyer, the useful idea is the split. The skill — the written, reviewable description of a task — is yours and can move. The connections, access, and context that let it actually run belong to whatever platform executes it. When you adopt a tool that supports this pattern, you keep the part that encodes your judgement and accept that the runtime around it may differ between hosts. That is a healthier dependency than one where the whole process lives inside a single product with no way out.
A vendor changing who runs it
Stability is a feature too. When you standardise a team on a tool, you are betting on the company behind it as much as the software. Leadership changes are a signal worth reading, not because change is bad, but because the direction of a product often follows the people steering it.
HubSpot announced that co-founder Brian Halligan is returning as Chief Executive Officer “to Lead a Refounding of the Company”, with Karen Burgess named as the next Chief Financial Officer [5]. The word refounding is the one to note. A founder returning to reset direction usually means priorities will shift, and if you run your business on a tool from that vendor, the roadmap you were planning around may move with them. This is not a reason to act. It is a reason to watch the next few product announcements more closely than usual, and to make sure your own data and workflows are portable enough that a change in direction is an inconvenience rather than a crisis.
What to take from the day
The common thread is that the AI badge on a product tells you almost nothing. What tells you something is whether the agent changes a number, who pays for the calls behind it, whether the work you build can move, and who is steering the company. Judge the four together and the marketing falls away.
Sources
- [1] Bring your own key (BYOK) — Linear
- [2] A Great B2B Agent Doubles Your ACV. Most AI Features Don’t Move It at All. — SaaStr
- [3] New Data: Holiday Shoppers Will Rely on Agents. Retailers Must Be Ready. — Salesforce
- [4] Build once, run on any agent: A practical guide to agent skills — GitHub
- [5] HubSpot Announces Leadership Transition: Co-Founder Brian Halligan Returns as Chief Executive Officer to Lead a Refounding of the Company; Karen Burgess Named Next Chief Financial Officer — HubSpot