On this page
Two columns, not one market
Remote work is not one market rising or falling together. It is two lists moving in opposite directions at the same time, often inside the same company.
On one side, the work AI already does well:
- Data entry and encoding
- Basic writing and rewriting
- Copy-pasting between systems
- Simple research and summarising
- Formatting, sorting, listing
On the other, the work that still needs a person:
- Seeing what is broken inside a business
- Building the system that fixes it
- Deciding who is allowed to see what
- Checking AI's work before a client sees it
These are not two separate industries. The property manager who used to hire someone to type enquiries into a spreadsheet is the same property manager who now needs someone to work out why enquiries go missing between the website and the sales team. Same business, same budget, different job. The demand did not leave. It moved along the list.
Why the shrinking column is shrinking
The usual explanation is that clients got cheap. That explanation is comfortable and wrong, and believing it keeps people applying for work that no longer pays.
Work loses its price when it becomes reproducible. A task is reproducible when it can be described completely enough that the description alone produces the result. Data entry qualifies: a defined input, a defined output, and a rule connecting them. Once that rule is written down, it runs again at almost no cost. From then on the price drifts towards the cost of running the rule, not the cost of an hour of your attention.
This did not begin with AI. Import tools, form validation and spreadsheet formulas had already taken the most mechanical parts. What changed is that AI extended reproducibility into work that used to need judgement about language: rewriting a paragraph, condensing a report, drafting a polite reply, pulling five facts out of a long document. Those tasks felt like thinking. It turned out that most of what they needed could be described.
There is a test you can run on anything you are paid for. Could you write instructions complete enough that a stranger who has never met the client could follow them and produce acceptable work? If yes, the task is reproducible and its price will keep falling. If the honest answer is "not really, you would have to know the client", you are holding something more durable.
That is what the published range of $4–7 per hour describes. It is not an insult and it is not the ceiling of remote work. It is the going rate for output a tool can now produce. A client weighing your hour against that tool is not being cruel; they are comparing two routes to the same result. Once you see it that way, the problem stops being personal and starts being something you can act on. What actually changed covers the shift itself in more detail.
The short version: the shrinking column is work you do. The growing column is work you are answerable for. Tools absorb the first. They cannot absorb the second, because responsibility does not transfer to software.
What makes the growing column defensible
Each of the four items resists description for its own reason. It is worth knowing which reason, because that is what you are defending when a client pushes back on price.
Seeing what is broken inside a business
The information this needs does not exist in writing. It lives in people's heads, in habits nobody documented, and in workarounds everyone learned and no one admitted to. Getting at it means asking careful questions of busy people who are slightly defensive about their own processes. You cannot hand a model something that has not been written down yet.
Building the system that fixes it
Building is a long run of small decisions with no single correct answer. Which information sits in which table. What happens when a staff member leaves. Whether a field is a free text box or a fixed list. AI produces the pieces quickly; someone still has to choose which pieces, decide how they connect, and live with those choices long after the build is finished. From prompts to systems walks through what that looks like in practice.
Deciding who is allowed to see what
Access rules are judgement about consequences, not configuration. A tool implements whatever rule it is given, including a bad one, and it does so quietly and correctly. It has no view on whether a contractor should be able to open the full client list, or whether finance figures belong on the same screen as the delivery schedule. Someone has to decide, and someone has to answer for it when the decision turns out to be wrong. Backend security when you are not a developer covers the thinking behind those rules.
Checking AI's work before a client sees it
Output that looks right and is wrong has become cheap and plentiful. A summary that quietly drops the one clause that mattered. A tidy list missing two rows. Whoever catches those things is carrying risk on the client's behalf, and carrying risk is a relationship rather than a task. It is the least glamorous item on the list and one of the easiest places to start.
The distance between those two columns is the whole argument for learning this. Read that as a description of the gap in the market, not a prediction about any one person's rate.
When you apply and hear nothing back
Sending applications and receiving nothing is a common experience in this market. It is usually read as a verdict. It is better read as information.
- The application is written in shrinking-column language. Detail-oriented, fast learner, proficient in spreadsheets, can handle admin and data entry. Every one of those phrases describes reproducible work, so the client prices it accordingly, if they read it at all.
- Nothing in it forces a second look. When many applications say the same thing, the client is not choosing the best one. They are scanning for anything specific and stopping there.
- The job may no longer exist. Listings stay up after the need behind them has been absorbed by a tool. Silence there has nothing to do with you.
What silence does not mean: that you are bad at your work, that clients are unfair, or that the answer is another certificate in the same skills. More proof of shrinking-column ability does not move you across.
No one can honestly tell you how long this change takes, what a client will pay you, or whether a particular application will get a reply. Anyone offering a timeline or a number is guessing. What follows is a change of direction, not a schedule, and no result is guaranteed by it.
What to change, in order
- Show one thing you built. A small portal, a working automation, a database with sensible access rules. A link a client can open answers the question your CV never can.
- Describe the problem, not the tool. "I built a system so the site team can log deliveries from a phone and the office sees them without chasing a group chat" tells a client what you are for. "Proficient in automation tools" gives them nothing to price.
- Stop answering listings priced for the shrinking column. A post asking for data entry at a fixed hourly rate has already decided what the work is worth. Applying to it more skilfully does not change that.
- Ask one question only an operations person would ask. What happens when two people update the same record? Who on the team should not be able to see the client list? A question like that is evidence in a way a claim never is.
- Offer to be the one who checks. Reviewing AI output before it reaches a client is unglamorous, immediately useful, and puts you on the answerable side of the line straight away.
None of this means abandoning what you already know. Someone who has done admin work for years understands where a business leaks time better than most outside consultants ever will. That knowledge is the raw material for the first item on the growing list. What changes is the use you put it to: instead of absorbing the mess by hand every week, you name it and build something that removes it.
Where to go next
If the map above matches what you have been seeing, these go deeper into the parts that matter most.
- The skills that outlast the tools — which of your existing habits carry across, and which do not
- What an AI business app actually is — the thing clients in the growing column are actually buying
- What the market pays — VA work versus AI-certified work — how the two columns show up in rates
- The curriculum — what is taught, session by session