Enterprises belief their AI brokers. They simply won’t let go of the leash.
That’s the clearest sign in 3,222 G2 evaluations by enterprise consumers of AI brokers in 2025 and 2026. Most of those consumers belief their agent. Only a few would let it act autonomously.
Consider a brand new rent who aces each job in week one. You continue to test their work earlier than it reaches a buyer. That’s how enterprises deal with brokers right now. We name them short-leash consumers. They’re enterprise reviewers who belief their agent “fairly a bit” or “very a lot” however wouldn’t give it complete autonomy. Their numbers almost doubled this yr.
TL;DR
- Enterprises are utilizing AI brokers extra. Enterprise evaluations of AI brokers on G2 greater than doubled, from 983 in 2025 to 2,239 in 2026. 53% of reviewers now use their agent more often than not or each time.
- Patrons belief their brokers however gained’t allow them to act alone. 77% of enterprise reviewers belief their AI agent, but solely 7% would give it complete autonomy.
- Setup nonetheless hurts. Setup and complexity are the highest complaints in enterprise evaluations, named in 25% of dislikes in 2026. Even reviewers who fee their agent 9 or 10 increase it.
- Integration impresses much less. Reward for integration fell from 30% of enterprise evaluations in 2025 to 21% in 2026, and it dropped in all 5 agent classes.
- Use is getting shallower. The typical share of customers who totally adopted their agent fell from 68% in 2024 to 55% in 2026. Almost 3 in 10 enterprise reviewers say 30% or fewer of the individuals meant to make use of the agent at their firm have totally adopted it.
High 5 agent classes with essentially the most reviewed merchandise
For our analysis, we checked out G2’s AI Brokers classes and selected the highest 5 with essentially the most reviewed merchandise, since product depend reveals the place distributors crowd in and the place sellers compete hardest.
Then we pulled each accredited assessment from firms with greater than 1,000 workers in these 5 classes, from January 2025 to October 23, 2026. That gave us 3,222 evaluations throughout 231 merchandise.
We learn the solutions throughout three themes:
- Scores: Probability to advocate, time to go reside, and person adoption.
- G2’s AI agent questions: Belief, autonomy, and what would increase belief, answered by 1,746 enterprise reviewers.
- Likes and dislikes: Tagged by theme with a key phrase methodology, so learn these shares as directional.
Merchandise counted from January 2024 to October 2026. A assessment can sit in a number of classes. Supply: Critiques on G2
AI isn’t simply including new classes. It’s altering how consumers decide outdated ones. Learn The Nice Software program Re-Ranking: High Classes Reshaped by AI in 2026 to see which classes consumers are rewriting.
1. Enterprises now use AI brokers extra, beginning within the again workplace
Enterprise evaluations of AI brokers greater than doubled this yr, from 983 in 2025 to 2,239 to date in 2026, in accordance with G2. G2’s 2026 AI Brokers Report backs this: 66% of organizations now run brokers in restricted or scaled manufacturing.
AI utilization is changing into extra of a behavior for this group, too. In 2026, 53% of enterprise reviewers say they attain for his or her agent more often than not or each time, up from 47% in 2025.
And the place do they use it? That occurs largely behind the scenes. Course of automation leads enterprise use circumstances at 34%, then customer support at 27%, and software program engineering at 21%. Gross sales growth (12%) and advertising and marketing (7%) path.
Evaluate highlights
- Nivas Simhadri, a system engineer at a retail enterprise, wrote in Might 2026 that one crew used Joule to automate journal posting and reconciliation solely. Automating buy approvals saved “a median of 70 hours of guide work per week.”
- Pradip Gadekar, a know-how lead at Infosys, wrote in April 2026 that ServiceNow AI Brokers “reduce down on guide effort” by assigning tickets and suggesting fixes on their very own.
How evaluations have modified
- In 2025, 47% of enterprise reviewers mentioned they used their agent more often than not or each time. In 2026, 53% do, and the share who use it each time rose from 11% to 17%.
What it means for software program sellers
Enterprises show brokers on inner work earlier than they belief them with income. That could possibly be why enterprise operations brokers grew so quick, at 3.5 instances from 2025 to 2026.
Sellers can promote into that sample. Anchor the primary deal to an operations job that the customer already measures, similar to lead routing or follow-up {that a} human approves. Then develop from there.
Or, if the deal stalls after a powerful demo, test what you requested the customer to belief first.
A front-office agent is a tough first step for a crew with no back-office wins but.
2. The short-leash purchaser trusts the agent however retains management
G2 has discovered that enterprise consumers largely belief their brokers: 77% belief them “fairly a bit” or “very a lot”. Reviewers have additionally answered how a lot autonomy they’d give the agent, and the solutions tighten quick:
- Autonomy with guardrails: 60%
- Restricted autonomy: 21%
- All the time have a human within the loop: 11%
- Whole autonomy: 7%
Supply: evaluations on G2
Right here is the half that shocked us: extra belief lengthens the autonomy leash, nevertheless it not often comes off. G2 asks reviewers how a lot they belief their agent to reliably carry out its supposed duties. As belief rises, consumers give the agent extra room.
Restricted autonomy falls from 34% amongst reviewers who belief it “considerably” to 10% amongst those that belief it “very a lot,” whereas autonomy with guardrails rises to 65%. However complete autonomy stays uncommon. Even amongst reviewers who belief their agent “very a lot,” solely 15% would grant it, and about 1 in 10 nonetheless desire a human to approve each motion.
Belief and autonomy reply completely different questions. Belief is about whether or not the agent does its job effectively. Autonomy decides whether or not anybody checks the agent’s work earlier than it acts.
With complete autonomy, a mistake can attain a buyer, a document, or a cost earlier than anybody catches it, and the customer’s crew is held accountable. That’s why the individuals who carry that danger maintain the leash tightest: 5% of admins and a pair of% of government sponsors would grant complete autonomy, towards 10% of on a regular basis customers.
So what would earn extra belief? 79% picked a visibility or management repair. They need the agent to elucidate its choices (32%), extra guardrail settings (29%), or a warning earlier than questionable actions (18%). Solely 21% requested for higher reliability.
Supply: evaluations on G2
G2’s AI ranking questions level the identical approach. About 500 enterprise reviewers scored their brokers from 1 to five on particular qualities in 2026. 92% gave output accuracy a 4 or 5, however solely 82% mentioned the identical for the way effectively the agent explains its outcomes or fixes its errors. Low or middling scores are greater than twice as frequent for explanations as for accuracy, 18% towards 8%. Most consumers are already glad that the agent will get the reply proper. What holds again their belief is seeing the way it acquired there and what occurs when it’s flawed.
Evaluate highlights
- An ideal rating doesn’t imply a hands-off rollout. The 2 reviewers under rated their agent 10 out of 10 and nonetheless in-built guardrails and a gradual path to extra autonomy.
- Monika Chouhan, a Salesforce Agentforce person at Cambridge College Press and Evaluation, wrote in September 2026 that “configuring guardrails, managing person permissions, and fine-tuning immediate templates require specialised Salesforce experience.” The assessment nonetheless provides Agentforce a ten out of 10.
- Byron Chen, a supervisor at Ontario Lecturers’ Pension Plan, wrote in April 2026 that it took “a while and a gradual rollout” earlier than the crew felt snug letting ServiceNow AI Brokers deal with extra requests. The assessment provides the product a ten out of 10.
How evaluations have modified
- Complaints about seeing why an agent acted rose from 3.0% of enterprise dislikes in 2025 to five.3% in 2026. Complaints about governance and management rose from 4.9% to six.4%.
Holding that management has a price. An correct agent nonetheless wants watching when consumers can’t see the way it reached a call, so somebody on the crew checks its work.
G2’s 2026 AI Brokers Report, a survey of B2B organizations, calls this the Mistrust Tax: 31% spend six or extra hours every week overseeing brokers, and 45% say oversight and rework eat up greater than 1 / 4 of the beneficial properties.
What it means for software program sellers
Cease main with autonomy. Solely 7% of enterprise consumers need it, and it worries the admins and sponsors who log out.
Lead with the leash as a substitute. Within the first demo, present how the customer units guardrails, sees why the agent acted, and will get warned earlier than a dangerous step. Make these controls simple to set, or the leash turns into a setup venture of its personal.
Fast tip: Put the invoice on the leash, too
In 2026, about 20% of enterprise evaluations of AI agent builders and AI buyer assist brokers on G2 complain about price, towards 9% for enterprise operations brokers. A number of describe usage-based pricing that’s laborious to forecast.
Earlier than the customer indicators, present them:
- What a typical month will price
- The place they’ll set spending limits
- How they’ll see prices as they construct up
Displaying these controls early additionally helps offers that stall in safety assessment, the place groups want proof they’ll see, steer and cease the agent
AI isn’t priced like software program, and consumers really feel it on each bill. Watch The right way to Management AI Software program Spend Earlier than It Spirals to know the price worries you’ll have to reply.
3. The early glow of AI brokers is fading in enterprise evaluations
Enterprise consumers have good issues to say about brokers, however there may be room for enchancment, discovered G2. The typical really useful rating in 2026 is 9.04 out of 10. Nevertheless it has slipped yearly, from 9.24 in 2024 and 9.10 in 2025. Excellent 10s fell under half for the primary time, from 55% to 48%.
Supply: evaluations on G2
4 of 5 classes slipped this yr. Three patterns clarify the slide:
-
Setup nonetheless hurts. Setup or complexity is the highest grievance in each years, named in 25% of 2026 dislikes. It peaks for assist brokers (34%) and SDRs (31%). Followers really feel it too: promoters increase it barely extra usually than different reviewers.
-
Integrations don’t impress consumers. In 2025, 30% of enterprise reviewers praised integration. In 2026, 21% did, and the drop confirmed up in all 5 classes. Our learn: integration is now desk stakes, so consumers cease mentioning it.
Supply: evaluations on G2
3. Adoption is shallow. The typical share of customers who adopted the agent fell from 68% in 2024 to 55% in 2026. Almost 3 in 10 reviewers now say 30% or fewer of their customers have totally adopted it.
Evaluate highlights
- Foram Chudasama, a guide at Tata Capital, wrote in July 2026 that the principle downside of Automation Wherever Agentic Course of Automation is “the training curve and implementation complexity.” The assessment nonetheless provides the product a ten out of 10.
- Ashfhaq Shaikh, a senior guide at Infosys, wrote in February 2026 that clunky screens, too many clicks, and chat interruptions stored Salesforce Agentforce at “low adoption (beneath 20%).”
How evaluations have modified
- Setup complaints rose from 23% of enterprise dislikes in 2025 to 25% in 2026. Reward for ease of use fell from 27% of likes to 22% over the identical interval.
What it means for software program sellers
A excessive ranking doesn’t imply the entire crew makes use of the agent. In 2026, enterprise reviewers on G2 rated their AI brokers 9 out of 10 on common for the way doubtless they’re to advocate them. However when requested what % of their customers have totally adopted the agent, they answered 55% on common, and almost 3 in 10 put it at 30% or much less. At renewal, the customer has to justify the price, and that case rests on how many individuals use the agent.
Sellers should fund setup contained in the deal, set an adoption goal for the primary 90 days, and observe it with the customer.
Retire the “it integrates with the whole lot” as a headline declare for agentic AI software program. Present what the combination lets the customer do.
Outline which outcomes you’ll show within the first 90 days
Right this moment, 7% of enterprise reviewers on G2 would give an AI agent complete autonomy, up from 6% in 2025, discovered G2 evaluations. At that tempo, it stays beneath 10% by 2027.
G2’s Chief Innovation Officer, Tim Sanders, expects belief to remain the toughest a part of the sale, and warns that crossing the chasm might be a bumpy trip. Patrons, he writes, need distributors to “present their work earlier than they take them off a leash.” His recommendation is to promote regular, repeatable beneficial properties in each day operations, give clients a transparent path to extra autonomy, and make an agent’s efficiency, prices, and controls simple to test.
So resolve now the way you’ll show success within the first 90 days. Present the guardrails consumers set for themselves, the choices the agent can clarify, and one slender back-office win they’ll measure. Show management early, and the customer will lengthen the leash.
The end result? Patrons usually tend to purchase your product or renew contracts.
Patrons now have a public method to test how an agent handles autonomy and coverage. Watch How G2 Agent Evaluations Rating AI Brokers for Software program Patrons to see the assessments your agent will face.
How a lot autonomy do enterprise consumers need to give an AI agent?
Little or no. Solely 7% of enterprise reviewers on G2 would give an AI agent complete autonomy. 60% need autonomy with guardrails, 21% need restricted autonomy, and 11% desire a human to approve each motion.
What would make enterprise consumers belief our AI agent extra?
Extra visibility and management. 79% of enterprise reviewers picked a visibility or management repair, led by clearer explanations of the agent’s choices (32%) and extra guardrail settings (29%). Solely 21% requested for higher reliability.
The place ought to we begin when promoting AI brokers to enterprises?
Within the again workplace. Course of automation is the highest enterprise use case at 34%, forward of customer support (27%) and software program engineering (21%). Gross sales growth (12%) and advertising and marketing (7%) path, so anchor the primary deal to an operations job that the customer already measures.
Ought to software program sellers nonetheless lead pitch with integrations?
Not because the headline. Reward for integration fell from 30% of enterprise evaluations in 2025 to 21% in 2026, and it fell in all 5 classes. Patrons now anticipate it, so present what the combination lets them do in week one.
How will we defend enterprise renewals after the deal closes?
Plan for setup and adoption from day one. The typical enterprise advocate rating fell from 9.24 in 2024 to 9.04 in 2026, and almost 3 in 10 reviewers say 30% or fewer of their customers have totally adopted the agent. Fund setup contained in the deal and set a 90-day adoption goal with the customer.
