What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Most large employers are treating generative AI like standard office software: buy licenses, train staff and track usage. Paysend CEO Ben Chisell is taking the opposite approach. He supports AI, but says employees should keep access only after showing that it creates useful work. New hires get roughly a month to test a tool, then must explain what it delivered before their access continues.
Why Paysend is not giving every employee an AI license
Chisell’s argument is not that AI lacks value. He estimates that Paysend can complete about 25% more engineering work with the same-sized team. That is his own 2026 estimate, not an independently audited productivity study, and it describes additional output rather than a promise of 25% higher revenue.
His concern is that inexperienced users can use AI to multiply weak habits. A polished draft, automated workflow or generated code can look productive while adding review work, errors or security exposure. “I know it sounds very unglamorous and unsexy as an approach given the world of tokenmaxxing and AI, but I think it’s really, really important,” Chisell told the CNBC report carried by KhanList.
The month-long trial
Paysend’s policy gives a new employee time to experiment rather than banning access outright. After about a month, the employee must explain what the AI produced and why the result justifies continued access. The test is therefore based on demonstrated value, not seniority or a permanent entitlement attached to a job title.
#1 Best Overall
What happened with an SEO lead
In one example, an SEO lead used a general-purpose Claude subscription to produce content briefs faster. Paysend judged that speed alone did not justify a broad subscription. Instead, the company developed a dedicated content-brief agent for the specific workflow. The distinction matters: an employee may be faster at a task without creating enough incremental value to cover the cost, risk and oversight of a general AI tool.
How this differs from broad corporate rollouts
Other employers are measuring AI adoption at much larger scale. The figures below are reported for 2026 and should be read as company-reported deployment or investment numbers, not as directly comparable experiments.
| Organization | Reported approach or result | What the figure does—and does not—show |
|---|---|---|
| Lloyds Banking Group | 40,000 Microsoft 365 Copilot licenses; 97% active use among license holders | Very high usage among licensed staff; it does not by itself establish revenue growth or net productivity. |
| Sage | Hundreds of thousands of work hours saved by AI coding tools | A reported time-saving total; the evidence provided does not state how the hours were measured or converted into revenue. |
| Grant Thornton UK | £500 million (about $660 million) AI investment | Planned or reported investment; spending is not the same as realized return. |
| Paysend | About 25% more engineering work with the same-sized team, according to Ben Chisell | An executive estimate of capacity, not an independent audit or a company-wide revenue result. |
A U.K. government study published in January 2026 illustrates the measurement problem. Three-quarters of AI-using businesses said workforce productivity improved, yet 77% said revenue was unchanged. Those are survey responses from businesses, not a controlled causal experiment. AI can help employees complete more tasks while sales, pricing, demand or capacity constraints prevent that extra work from becoming revenue.
Rank #2
Does giving everyone AI improve revenue?
There is no automatic link. A license can increase activity—more prompts, generated documents or completed code—without improving the outcomes a business pays for. Revenue may remain flat when additional capacity is used for internal work, absorbed by quality checks, or left unused because the organization has no new demand.
The useful comparison is not “everyone versus nobody.” It is whether the company can connect access to a measurable business result:
- Output: completed engineering tickets, accepted code, qualified leads, resolved cases or other work that meets a defined quality bar.
- Quality and risk: defect rates, rework, security findings, policy violations and customer-impacting errors.
- Time to value: cycle time from request to an accepted result, rather than time spent generating a first draft.
- Capacity use: whether saved hours fund additional growth work, improve service levels or simply reduce staffing needs.
- Cost: license, model, integration, training and review costs compared with the value of the accepted output.
Why faster can feel like productive
CoreView CEO Simon Azzopardi describes a failure mode that applies to both broad and selective access: “You feel more efficient, but you actually aren’t.” He was referring to using AI to build outside one’s expertise. A system can produce a plausible result quickly, while the user lacks the knowledge to detect flawed assumptions, insecure code or an unsuitable design.
CoreView’s agent-heavy development experience suggests that coordinating multiple agents can compress timelines when permissions, workflows and technical review are well controlled. The same model becomes risky when agents are allowed to make decisions in areas the responsible person cannot evaluate. Cycle-time reduction counts only when quality remains stable and someone can own the result.
Free tools Windows power users keep installed
One-click scans. No signup required.
What a responsible AI access policy should include
Set an evidence-based trial
Give employees a defined evaluation period, such as Paysend’s roughly one month, with a written task, baseline and success criteria. Require a short explanation of what changed, what the AI contributed and what review was needed.
Match permissions to expertise
Separate drafting or analysis access from permissions to deploy code, contact customers, change records or trigger transactions. An agent should not receive authority merely because it can perform a task quickly.
Measure accepted outcomes
Track completed work that passes review, not prompt counts or active-user percentages alone. Compare quality, rework and cycle time with a pre-AI baseline and state whether the measurement covers a team, project or individual.
Keep a named human owner
Chisell summarizes the governance principle this way: “Having a human in the loop shouldn’t mean human CC’d. It should mean human ownership.” The employee or manager responsible for the outcome remains accountable even when an agent performs intermediate steps.
Should companies limit AI to top performers?
Limiting access solely to a permanent “top performer” category would create its own problems: unequal opportunity, hidden experimentation and a risk that managers confuse confidence with competence. Paysend’s approach is narrower and more testable—access is earned by showing a useful result, and a specialized tool can replace a general subscription when that is the better fit.
Best Value
Companies also need to decide what happens to the capacity AI creates. Chisell says, “I personally believe you should use it to expand your ambition rather than make the team smaller.” That means setting a plan for redeploying saved time into new products, faster service or higher-value analysis, while being honest if the actual objective is cost reduction.
A practical decision rule for executives
- Define the business outcome before buying licenses.
- Run a time-limited pilot with representative users and a documented baseline.
- Give each user only the data and system permissions required for the task.
- Review accepted output, quality, rework, risk incidents and total cost.
- Continue, expand, redesign or withdraw access based on those results.
Broad licensing can be sensible where workflows are standardized, training is strong and outcomes are measured. Earned access can be better where misuse is costly or a specialized agent can outperform a general tool. In either case, high usage is an adoption statistic—not proof that AI improved the business.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errors
