The aibl Download: 7 Links to Close Out and Look Ahead

From real AI costs to human resistance: seven sharp reads on moving from pilots to production in 2026.

The aibl Download: Links to Close Out 2025

2025 proved agents work in production. The question for this year is less about what’s possible than how we operationalise AI since it sits in a new management space that’s half employee and half tech stack.

This edition starts the new year with seven favorite articles and threads about what works, costs to anticipate, people problems and the approaches you need to move from watching to doing.

1. The True Costs of AI Implementation

A refreshingly honest breakdown of what AI projects actually cost for UK SMEs in 2025. The bulk of the cost lies below the waterline in infrastructure and data hygiene. Data Prep is running 40-60% with another 15%-20% for training.

The aibl Takeaway: The era of cheap experiments is over. If a vendor quotes you £50k for a project but ignores the data audit and change management, the real cost will likely be double. Use this report to sanity-check your 2026 budget. 

2. Mid-Market CFOs Have The Hardest Job in AI Adoption

With 90% of CEOs now expecting ROI within a set timeframe (up from just 22% a year ago), the pressure is on to “bring the receipts.” This can lead to catastrophic errors.

A $500M logistics firm cut 15% of its staff anticipating AI efficiencies. The AI wasn’t ready. The result? Morale dropped 30%, remaining employees were overloaded, and retention tanked.

The aibl Takeaway: Don’t promise the board efficiencies the system can’t yet produce. AI is likely to require greater headcount (or at least talent cost) before it offers this type of ‘gain.’ But the metric for AI success isn’t payroll reduction; it’s revenue growth per employee. 

3. The Cost of Manual Drag

Research shows UK invoices cost between £4–£25 each to process manually. AI reduces that to pennies. For a business processing 200 invoices a month, that’s up to £36,000 a year. That’s a fair start to your AI budget right there.

The aibl Takeaway: While every leader should be thinking ahead and reimagining their business, today’s gains are in greasing the wheels of manual, repetitive work.

4. The Psychology of AI Resistance

We love this one because it’s so very human. A study of nearly 29,000 software engineers revealed a “competence penalty” where employees who visibly used AI were rated as 9% less competent by their peers, even when the work quality was identical. For women, the penalty was 13%.

The aibl Takeaway: To some mid-market employees, using AI can look like cheating or laziness. If you don’t offer “private mode” sandboxes where people can learn without judgment and attack this attitude as a cultural imperative, your expensive AI licenses will gather dust. Stop public leaderboards. 

5. Resistance is a Mirror

If your team is fighting your new AI workflow, it’s very likely to be “emotional uncertainty”. We just wrote a report on this very topic (coming next week). The common mistake is responding with logic and data when the root cause is emotional. Some of the culprits:

• The Silent Freeze: They don’t feel safe speaking up.
• The Loud Pushback: They care but don’t feel heard.
• The Passive Stall: They are overwhelmed and fear failure.
• The Quiet Underminer: They fear losing control or influence.

The aibl Takeaway: Change is threatening unless you’re the one helping to drive it. It’s been true in every wave of innovation and AI is the mother of all digital transformation. A happy, engaged workforce is the single most powerful determining factor in any corporate reinvention.

6. When an Employee Uses AI Too Much

A manager laments a direct report who runs every single task through Claude, stripping away all context, nuance and human judgment, producing over-engineered, even inappropriate work.

The aibl Takeaway: We need to delegate the process, not the purpose. AI is a tool to augment our thinking, not replace it entirely. 

7. How to Build Agentic Workflows (Without Coding)

Most people confuse rigid “AI Workflows” with adaptive “AI Agents,” leading to brittle systems. The winning model is a hybrid: an “Agentic Workflow” where a reasoning layer (the agent) decides which rigid, reliable workflow to trigger.

The aibl Takeaway: If you can’t articulate your business process with the clarity of a standard operating procedure, no agent can automate it for you. The bottleneck isn’t the AI model; it’s the quality of your own thinking.