New G2 Research: AI Agents Have Moved From Pilot to Production. Now Comes the Distrust Tax.

No software innovation has received more hype in recent years than AI agents, and the promises have been epic: They can make us 10x employees, accelerate workflows into competitive moats, and unleash business growth without adding headcount.

But in 2026, the AI agent market faces a bumpy road. Visionaries funded pilots last year with discretionary budgets and "AI war chests," yet those early adopters are only a fraction of the market vendors promised investors they could capture. The largest group of buyers - including both business process owners and IT - is wired for proof, not promises.

Strategist Geoffrey Moore named this moment in his book, "Crossing the Chasm." He wrote that the point of greatest peril in a high-tech market is the transition from an early market of a few visionary customers to a mainstream market of pragmatists. Released today, G2's 2026 AI Agents Report: Crossing the Agentic Chasm - based on a survey of more than 1,000 B2B software buyers and G2 review data - shows where agents are earning the trust of those pragmatists and where they are not.

Key findings from G2's 2026 AI Agents Report

Our research reveals that agents are being deployed faster than many companies can manage them, while also uncovering what B2B buyers now expect from agent vendors.

1. AI agent adoption is advancing from pilots to production

Companies are putting agents into the workflows that have real business impact, particularly where performance directly affects speed, quality, and cost. AI agent adoption by B2B software buyers climbed by 25% year over year, and 66% of organizations now have agents in limited or scaled production.

Nearly half of organizations with completed pilots say the majority of their pilots have advanced to full production. The next hurdle is giving agents more consequential work without adding enough supervision to erase the gains.

2. The Trust Gap has spawned the Distrust Tax

In our 2025 Agents Report, we described the Trust Gap. In 2026, it didn't disappear in production, but moved inside the workflow and became the Distrust Tax - the ongoing cost of running agents you don't fully trust.

Agents can create meaningful speed and cost gains, but companies often give part of that value back through reviews, corrections, exception handling, and oversight. 31% of B2B buyers say employees spend 6 or more hours each week bot-sitting (overseeing agent work). 45% say time spent overseeing and correcting agent mistakes consumes more than a quarter of the productivity gains they expected, and 46% report higher-than-expected operating costs. The longstanding habit, "trust but verify," comes at a steep cost.

3. Shadow agents are the new Shadow IT

As agent populations grow, inventory, identity, visibility, and accountability are lagging. 38% of organizations with agents in production operate more than 100 agents, yet almost half (47%) still use informal or decentralized management models. With agents operating at scale, companies need better visibility into which agents their employees are running, the jobs they are doing, and who answers for their work.

4. Agents are disrupting SaaS

Agents are reshaping the way buyers value SaaS. Seat-based pricing fits poorly with the work that agents perform continuously, and only 9% of buyers consider per-user or per-seat pricing fairest. Meanwhile, 72% say a strong ecosystem of connectors, actions, skills, or marketplace offerings is very important or a must-have for agent adoption. SaaS isn't going anywhere, but by 2030, 56% of agent buyers expect agents to orchestrate work across their SaaS stack.

5. Buyers hold agents to a higher standard of proof

Agents are capable, and they want autonomy. However, buyers want the vendors behind them to show their work before they take them off a leash. 52% require human quality review before production, and 77% re-evaluate agent performance at least monthly. As agents take on higher-stakes work, that evidence must hold up in the real world.

What these report findings mean for agent vendors

1. Sell repeatable gains, not moonshots: Among teams running agents in mature workflows, 4 in 5 report cycle-time gains of at least 11%. Pragmatists don't need a 10x promise. Show the measurable improvement, what it means in the customer's operating model, and where the same proof applies next.

2. Make the cost of oversight visible: Show buyers how approvals, review, exceptions, and rework affect net cycle time and cost per completed outcome. Then give customers a clear path to greater autonomy by defining what an agent can handle on its own today, when it should escalate, and which thresholds allow it to take on more. Build the operating model into the product too: Trusted data, integrations, guardrails, recovery paths, and implementation support are what turn a promising deployment into a repeatable business capability.

3. Make every agent governable: Automatically create and continuously update a record for each agent, including owner, workflow, identity, permissions, connected systems, activity, and cost. Treat observability and recovery as core capabilities, with live activity histories, policy alerts, escalation paths, and one-click controls to pause an agent or revoke its access. Then use open APIs and ready-made integrations to feed that data into the systems IT already uses, so customers don't have to govern each agent through a separate dashboard.

4. Let customers shape pricing and the ecosystem: Test subscription, pay-as-you-go, and outcome-based options rather than forcing one model across your customer base. Follow the signals from adoption, usage, expansion, and retention, and be willing to move from annual recurring revenue to annual revenue per account as the market matures. Make agent readiness concrete by productizing the connectors, actions, APIs, and data context that let agents work across the systems customers already use. Then make handoffs, permissions, data exchange, and reliability across other products as strong as the experience inside your own.

5. Prove performance beyond the demo: Publish a testable definition of performance that shows the tasks, operating context, success criteria, policy constraints, and failure modes behind your claims. Give buyers testimonials, evaluation reports, reliability data, and references, along with a clear explanation of how the agent behaves when the workflow gets messy. Then treat independent evaluation as part of the product story: Pragmatic buyers need credible, comparable signals to verify performance before and after they give an agent more control.

Proof is what crosses the chasm

The age of agents is upon us. The market is booming, and agents will ultimately revolutionize our systems of work. The vendors that cross the chasm will be the ones that make performance legible, costs manageable, and control credible.

But buckle up. It's going to be a bumpy ride across the chasm.

For more insights, read the full report, G2's 2026 AI Agents Report: Crossing the Agentic Chasm.