The ‘false confidence’ trap and what it takes to make AI actually change the business

The ‘false confidence’ trap and what it takes to make AI actually change the business

Plus: The two-week fix that tripled leads for a sales agent

From the aibl team

Teams work hard to get AI budgets approved, but then nothing deploys because middle managers slow things down, which looks like resistance to leadership.

Sound familiar?

At last week’s aibl Advisory Board, founders and operators landed on a different read: the hesitation is rational.

Leadership approves tools but doesn’t adjust what gets measured. They believe the AI can work miracles, but it can’t. They want experimentation but don’t carve out time for it. They want to redesign without changing what performance means. If that’s the setup, caution from the middle isn’t obstruction, it’s responsible.

The article below walks through what our Board members are seeing and the moves that work. It covers changes to mandates, how teams are paired and what performance means when building capability. It’s worth reading if this feels familiar.

Your biggest AI problem isn’t resistance – it’s false confidence

Matt Shumer’s viral post last week argued the gap between what AI can do and what most people think it can do is “enormous and dangerous.” He’s right. As our CEO Richard Breeden commented, Shumer is “right(ish)” about the speed and scale. What would have taken days and cost thousands is now nearly instant, zero-cost, and almost as good.

That speed is what makes the risk Shumer pointed out so dangerous. People who tried AI a year or two ago got underwhelming results and concluded it’s not relevant to their role. The same applies to those using unpaid and lower end LLM versions today in comparison with the top models today. Their impression is stuck on a reality that no longer applies to those who have kept up with advances and are taking advantage of the best in class.

But the gap runs both ways. We keep seeing leaders who engage just enough to feel “done”. The question isn’t whether AI works. It’s why so many companies think they’re getting so little from it. In most cases, the use is real and frequent, but basic and optional — so the business doesn’t change.

News worth reading

Don’t blame the tech, AI alignment is a leadership problem

The false confidence described above shows up clearly in Zellis’s new research. 94% of leaders say their organisation uses AI, but only 61% of employees actually use it in their role. Leaders are more likely to want AI applied to high-stakes decisions like pay and promotions. Employees want it for routine admin. That gap isn’t a communication problem — it’s a structural one.

The most telling finding is about involvement. 63% of leaders say they involve employees in AI decisions. Only 40% of employees agree. When a third of your workforce actively disagrees that they’re part of the conversation, adoption numbers on a board deck mean very little.

Only 45% of the wider workforce believes senior leaders are using AI effectively. That tracks with what we see across aibl’s work with mid-market teams — the gap between leadership perception and workforce reality is where most adoption stalls.

What sales automation looks like when it works

Last week, we wrote about agents that succeed by integrating with tools teams already use. This week, a mid-market property management founder showed us what that looks like in practice.

Landlords shopping for a management company often contact several firms at once. The one that responds fastest with the right first question usually gets the reply.

His sales team was losing 40–60% of inbound leads to response time. This was especially true for enquiries during peak periods and after office hours. Government reforms had reduced the supply of potential landlords, so growth through volume wasn’t an option anymore. He needed to convert better.

After reading our newsletter on agent integration, he spoke with his engineering team. They settled on two AI agents, each tied to a specific gap in the sales workflow.

An inbound agent monitors enquiry forms and emails, responding in under 60 seconds with a qualifying question. That covers the after-hours gap where most leads were dying. An outbound agent researches prospects and sends personalised sequences. It gives the sales team a proactive pipeline they didn’t have time to build manually.

Both agents had clear handoff points. The inbound agent either offered a booking link or flagged threads for a sales rep to pick up. The outbound agent identified warm prospects for the team to close. The sales team still handled the actual conversations. The agents covered speed, consistency, and follow-up.

Product spotlight of the week

Last week we saw Mindstone demo Rebel, their AI agent platform aimed at mid-market firms. It’s designed to answer the question that’s holding back most leadership teams: how do you let AI act autonomously without it doing something catastrophic?

Rebel connects to your work tools — email, calendar, Slack, CRM, and document systems — and executes multi-step tasks. In the live demo, it built an investor pitch deck, researched dozens of attendees in the room, drafted personalised emails to each based on that research, then queued them for sending. All of that happened in parallel while the founder was still presenting.

The system maintains two types of memory. Company memory holds shared context — active deals, current priorities, who’s working on what. Personal memory tracks your work style, your projects, your preferences. So when an agent acts, it has full context without you repeating yourself.

For operators impressed by fully autonomous agent platforms but nervous about deploying them without safeguards, Rebel answers that with selective approval. The bulk of the work — research, drafting, synthesising, coordinating — runs autonomously. You only approve at decision points: sending external emails, sharing sensitive data, modifying important documents. You’re steering, not babysitting.

At the event, Mindstone shared that Epignosis — a 250-seat learning management software company — has committed €250k to embed Rebel across every function, expecting multi-million return. Their CEO worked with Mindstone for a year on practical AI adoption before committing to go fully AI-native company-wide. For mid-market operators, that makes them one of the early test cases for whether agentic AI can prove it works at scale.

Copyright © 2026 aibl. All rights reserved.
aibl, Office 318, X Arding & Hobbs Building, 7 St. John’s Road, London SW11 1QN