New G2 Research: AI Agents Have Moved From Pilot to Production, But Oversight Comes at a Cost

G2’s survey of 1,019 B2B decision-makers found AI-agent purchasing rose 25% year over year, from 69% to 86%, and 66% of buyers now have agents in limited or scaled production. Adoption is producing reported workflow benefits, but 31% spend at least six hours weekly overseeing agents and 45% say oversight and rework erode more than a quarter of expected productivity gains. Governance remains informal at 47% of organizations, while 84% say agent capabilities will influence their 2027 software choices.

Analysis

The investable signal is not simply stronger AI demand; it is a shift in where software value accrues. As agent fleets move into production, buyers are likely to favor vendors that make agents governable across existing systems—identity, monitoring, audit trails, workflow orchestration, and connectors. That raises the strategic value of control planes and integration ecosystems, while increasing the risk that standalone agent features are bundled into incumbent platforms.

The pricing shift creates a second-order risk for SaaS economics: task- or consumption-based billing aligns price with realized work, but can weaken predictable seat expansion and expose vendors to usage volatility. It may also transfer productivity gains to customers rather than expand vendor margins. Model providers are not automatic winners: inference expense, exception handling, and human review can constrain customer ROI and limit willingness to scale usage.

Treat G2’s survey as a directional demand indicator, not proof of durable ROI. It is self-reported, and G2 has a commercial interest in software evaluation; validate against vendor guidance, net retention, usage, and customer-reported payback. Near term (days), the report alone is unlikely to justify a broad AI-software rerating. Over 1–3 months, watch earnings commentary for agent attach rates, consumption revenue, and support costs. Over 6–18 months, governance and cross-stack orchestration may become procurement requirements. The thesis weakens if agent deployments stall, review costs remain high, or vendors fail to show incremental revenue without margin dilution.

Key Decisions for Investors

• Do not chase a broad AI-software basket on this survey. Treat it as a catalyst to prioritize diligence, not as evidence of realized earnings growth.
• Build a relative-value watchlist: favor vendors with credible identity, observability, governance, and cross-platform integration capabilities over products differentiated mainly by standalone agent features. Confirm agent-related revenue contribution and implementation economics before taking single-name exposure.
• Monitor seat growth and net retention at seat-priced SaaS vendors alongside consumption revenue, inference costs, and support expense. A rising agent usage metric without improving retention or gross margins would argue against paying for the theme.
• Reassess after the next earnings cycle: upgrade the thesis only if companies disclose repeatable production deployments, measurable customer payback, and agent monetization that offsets any seat compression; reverse it if deployments remain pilot-heavy or operating costs outpace agent revenue.