Who Should Pay for Personal AI at Work?

A CAMO engineer pays for a ChatGPT plan on her own card. On Monday she uses it to tidy an email to a lessor. On Tuesday she asks it to summarise a service bulletin. On Wednesday she drafts an AMP review note with it. By Friday it has done more work for her employer than some of the software the employer actually pays for.

She is not unusual. A Part-147 instructor uses Claude to turn lesson notes into quiz questions. A quality manager uses Perplexity to research a regulatory change before a management review. A freelance auditor keeps one subscription for every client.

The account is personal. Much of the value goes to the employer.

That raises two questions. Who should pay? And is a general chat subscription the right tool for the work it is quietly doing?

Personal plans are quietly becoming work tools

Personal AI plans are sold as individual productivity tools. In practice many of them are work infrastructure, used during working hours for serious tasks:

• drafting emails to customers, lessors and authorities;
• summarising reports, manuals, contracts and meeting notes;
• preparing presentations, procedures and training material;
• researching technical and regulatory questions;
• rewriting findings and internal communication;
• producing first drafts of analyses and recommendations.

The employee pays. The employer gains speed and polish. The company often cannot see how much of its work passes through that account, and the user often has no guidance on what can be pasted, which sources to trust or who reviews the output.

So the tool can sit at the centre of someone's working day while being outside the company's process, controls and budget. The shadow AI article covers the risk side of that in detail.

Who should pay?

The honest answer depends on who the work is for.

If you are employed and the tool does regular work for your employer, it is reasonable to ask the employer to pay. Not only for the money: once the organisation pays, it knows the tool is in use and can set rules about data, sources and review. An unpaid personal plan doing company work is invisible, and invisible tools do not get policies.

If you are a contractor or consultant, the tool is part of your own kit, like your laptop or your reference library. You will usually pay for it yourself. The question then is not reimbursement but whether the tool is fit for the client work you use it on, and whether your clients' data is allowed to go into it.

If the use is mixed, which is the most common case, the split in the next section is where most people end up.

Company workspaces solve one problem and create another

More employers now provide team plans, enterprise workspaces or controlled AI environments. That fixes a real problem: the company gets control over access, data handling, billing and policy.

It also creates a new split for the person.

A company workspace belongs to the employer. It is not the place for personal questions, private research, career planning, study for a licence module, or preparation for the next job. So the same engineer ends up with two AI lives:

• a personal account for private work;
• a company workspace for employer-controlled tasks.

That separation is right. Work data should stay in work systems, and personal material should not live in company-controlled spaces.

But it shows the limit of the subscription model. The tools are split by who owns the account, not by the kind of work, the level of risk or the structure the work needs. A company workspace solves billing and data control. It does not, by itself, make a general chat tool fit for regulated work. And it stays behind when the person leaves.

Generic tools are strong, but the structure depends on you

ChatGPT, Claude and Perplexity are good because they are general. You can ask for a draft, a summary, a comparison, a checklist or a critique, and switch topics in seconds.

Aviation work, though, has requirements a blank chat box does not enforce:

• which authority and which Part apply;
• which source is authoritative, and whether it is Regulation, AMC or GM;
• which amendment is in force;
• what the organisation's approved exposition says;
• which format the output must take, such as a finding, a CAP or a compliance matrix;
• who reviews it before it is used.

A careful user can get much of this from a generic tool by prompting well, supplying the source text and checking every answer. That is the weakness: it all depends on the user remembering every time. A strong prompt on Monday and a rushed one on Friday produce very different work, and nothing in the tool notices the difference.

A wrong answer weighs more at work

A wrong answer about a recipe wastes an evening. A wrong answer in an audit finding, a procedure draft or an AMP review note can end up in a record, in front of an auditor or in a decision about an aircraft.

The risk is highest just outside your own specialty. A Part-145 quality engineer will spot a bad answer about Part-145. The same engineer asking about Part-147 training obligations, or a CAMO question from a lessor, is working next to their expertise, where a fluent answer is harder to catch. The structure looks professional. The tone sounds sure. The basis may be missing.

That is not a reason to stop using general AI. It is a reason to match the control to the work. The generic vs aviation AI article sets out what that control looks like, and the ChatGPT for EASA compliance article gives a verification routine you can use today.

What fits which job

Not all AI use carries the same risk, and it does not all need the same tool.

Writing a birthday message, drafting a management review summary, interpreting a regulation and preparing a safety procedure do not carry the same risk. Treating all of them as "just ask the chatbot" is the mistake.

Whatever mix you use, an organisation should write down what may be pasted into which tool and who reviews AI-assisted output. The aviation AI policy article covers what that policy should define.

Where Avioverse fits

Avioverse is a personal aviation workbench for individual professionals in an EASA environment. It is built for the third row of that table.

• It belongs to the professional. Your certificates, notes and working context stay with your account when you change employer.
• An employer can still pay. A company can pay for your subscription with a company card. Your personal records stay yours; work you do in an organisation's team workspace stays with the organisation.
• It is aviation-first. Metis, the assistant, works from a built-in EASA regulation library, labels what it quotes, and shows whether an answer was checked against sources.
• The human decides. Metis prepares drafts; you review, approve and enter official work in the correct company system.

Pricing is on the pricing page: Free to start, and Pro in three usage sizes with the same features. For a side-by-side view against general AI subscriptions and spreadsheets, see the comparison.

The real question

"Who pays for AI?" matters, and the answer is often "the employer, if the employer benefits".

The better question is which kind of AI system fits the work. Personal plans are good for personal thinking. Company workspaces are good for controlled business use. Aviation work that has to be sourced, structured and reviewed needs a tool built around those steps, not just a better prompt in a general one.