Pricing for uncertainty in an agentic world

The tech world is ablaze with speculation about a future in which work is largely done by agents, where agents become the buyers and users of software. It's an exciting turning point, and we're approaching it faster than many thought possible. But we often gloss over what happens between now and then: the awkward middle when teams are still figuring out exactly how to design agentic businesses.

But that "awkward middle" can't be ignored, because it defines how we do business right now.

The truth is that we're still figuring out exactly what work is meant to be done by humans and what we trust to go to agents. That answer looks a little different for every team. And it changes constantly as models improve and products evolve. And within each team, the answer keeps moving as pilots expand, stall and expand again.

So we're facing a new challenge: how do we price in a way that honors both the excitement around that agentic future and the uncertainty it carries? Flexibility has always been fundamental to usage-based pricing. Customers could flex how much they used, and with commitments, when they used it. Now we have to model a new dimension of flexibility: who, or what, is doing the work.

Human-agent teams are here, and their impacts are scaling as models get smarter

Agents have moved from demos into the org chart. In Microsoft's 2025 Work Trend Index, 46% of leaders said their organization already uses agents to fully automate workstreams or business processes. The same report coined a name for the people running them, the agent boss, and found that leaders expect their teams to be training (41%) and managing (36%) agents within five years.

Product teams are building for companies in transition

Building for this moment asks product teams to hold two things at once. Agents take on more of the work every quarter, so products have to give them real jobs: work they can pick up, run on their own and finish. But people still own the outcome, so the same products have to make it easy to check what an agent did, hand work back and step in when it gets something wrong.

In other words, the products getting this right make the handoff cheap in both directions, from person to agent and back again.

GitLab and HubSpot are two of the clearest examples. In both, people and agents work in the same place, and work can pass from one to the other without anyone changing tools.

GitLab lets you assign work to an agent the way you'd assign a teammate

In the GitLab Duo Agent Platform, a flow is a set of agents that runs a multi-step workflow in the background. You start one the way you'd hand work to a colleague: assign it to an issue or merge request, add it as a reviewer, or mention it in a comment. The person's job is to trigger the flow and review what comes back. So a developer can take an issue one week and hand the same kind of issue to an agent the next. (See GitLab in the Pricing Atlas.)

HubSpot puts agents on the front line and routes the hard cases to people

HubSpot's Agent Hub runs agents across marketing, sales and service. The Customer Agent resolves common support questions around the clock and leaves the team free for the cases that need a human, and HubSpot reports a human handoff rate right next to the resolution rate. On the sales side, reps approve the CRM updates and follow-up drafts that agents prepare after each meeting. (See HubSpot in the Pricing Atlas.)

Most AI pricing puts too much pressure on buyers to have the answers

Hybrid pricing has become one of the most popular ways for companies to sell AI products. It grew out of the need for SaaS companies to capture new revenue opportunities without sacrificing their core business, which is structured around seat-based value.

This traditional SaaS pricing is based on certainty - buyers understand what their budget is, what their seat count will be, and the exact problem a platform is meant to solve. But AI and consumption-based pricing broke this script.

In hybrid pricing, users continue to pay a seat-based fee, but unlock the flexibility to try consumption-based elements and pay only for what they use. It allows users to try these new AI tools without having a pre-defined idea of how much they needed or how valuable they would be, leaving more room for uncertainty.

Later, as these companies began selling consumption-based products into the enterprise, they butted up against a new challenge: companies that had certainty about the budget but uncertainty about exactly how and when it might be applied. So usage-based companies began modeling this with credit blocks or prepaid commitments. Companies could commit up front to a level of spend, and apply it to different consumption based products at different rates throughout the year, billed monthly with true-ups.

Now, companies are forced to price for a new dimension of uncertainty: who will do the work? Do we need more budget for humans or agents? Buying seats and tokens up front forces companies to answer a question that is rapidly becoming one of the biggest points of experimentation for companies at the frontier.

How companies are pricing for the future, and the transition

GitLab and HubSpot have both shipped pricing that lets customers change the mix as they learn. They start from different places and land on the same idea.

GitLab Flex lets teams plan one budget and decide what it buys

GitLab Flex is one annual commitment that covers seats and GitLab Credits, the $1 currency behind GitLab's AI features. Customers lock in a rate up front, re-allocate the spend between seats and credits as needs change, and track it against an annual balance. Each month's mix is reserved before the 1st.

For a team making the transition to agent-augmented work, this means planning around one number and flexing inside it. The team can add people one quarter and lean on agents the next, without reopening the contract. GitLab also offers Flex on every paid plan, so the teams most likely to be in flux don't need enterprise leverage to get it.

1. One commitment covers people and agents. Seats and credits draw from a single annual number, so the team plans one budget.
2. The mix can change mid-term. Spend moves between seats and credits as the team learns what agents can take on.
3. Some features are already priced for both. GitLab Orbit sells query packs that form one shared pool for humans and agents.

HubSpot Credits pairs the swap with governance that protects the budget

HubSpot Credits sells one rate, $10 per 1,000 credits, usable anywhere in HubSpot. Seats to Credits Swaps lets customers move spend from seats to credits. Around it, HubSpot hands customers the controls that usually sit in the vendor's back office: spend caps per feature, usage reporting and feature pausing. Customers choose credit packs up front or pay as they go.

The controls are what make the swap usable for teams navigating agentic transformation. A spend cap lets a team try an agent with a known worst case. Feature pausing is the way back when a rollout stalls. HubSpot launched all of this on its own page inside Agent Hub, and gave the controls as much space as the price.

1. Spend caps per feature. A team can try an agent knowing the worst case up front.
2. Seats to Credits Swaps. Budget moves from people to agents without a new contract.
3. Feature pausing. The way back when a rollout stalls, with credits shifted to the tools driving results.

Letting customers change the mix takes billing that can keep up

Most billing systems treat seats and credits as separate products, each with its own meter and its own invoice line. A shared budget puts them on one balance that the customer can rebalance at any time. A few things have to be true for that to work:

1. The exchange rate is set on purpose. Once budget moves between seats and credits, the conversion rate is a price. It moves your margin too, since seat revenue carries software margins and credit revenue carries inference cost.
2. The drawdown rules are written down. One account can hold included credits, purchased credits and a commitment at the same time. Customers need to know which burns first and what happens when a seat leaves mid-month.
3. Revenue recognition follows each dollar. Seat revenue is recognized over the term and credit revenue as it's used, so every dollar that changes form needs a record your ERP can follow.
4. Controls run on fast infra. Agents can burn through a budget in minutes, so metering, spend caps and entitlements have to update as usage happens. A spend cap checked against yesterday's usage lets an agent overspend today.

Price for the transition, because the answer will keep changing

It's tempting to design pricing for the destination, where agents do most of the work and outcomes are the main meter. That world may arrive, but customers will spend years getting there, one workflow at a time, with plenty of reversals along the way.

For now, the most honest answer to "who's doing the work?" is "we're still figuring it out." The pricing that wins will be the pricing that lets customers say that out loud.

To see how 167 companies price seats, credits and everything in between, browse Orb's Pricing Atlas, including annotated walkthroughs of GitLab and HubSpot. If you're modeling a move like theirs, Orb's prepaid credits, spend controls and simulations are built for it.