The data engineering market feels uncomfortable. Freelancers worry about their next contract. Experienced developers see companies replacing expensive employees with cheaper alternatives. AI appears in almost every conversation about reducing development costs.
The uncertainty is real. But if your response is that AI produces terrible code and someone will always have to clean up the mess, you may be building your career strategy around an outdated experience.
My opinion is straightforward: people who sell mainly coding capacity face growing pressure. Experienced professionals who use AI to solve customer problems have a substantial opportunity.
For freelancers, adapting starts with understanding that distinction.
Stop judging today’s tools by last year’s experience
What many people called “vibe coding” a year ago looked like this: ask a chatbot for code, copy the answer into an editor, discover that something was broken, and repeat.
That workflow still exists. But capable coding agents can do much more.
Tools such as Claude Code can inspect an existing repository, follow dependencies across files, implement changes, execute tests, examine failures, and revise their work. They can participate in an iterative development process instead of producing an isolated answer.
Yet the criticism often remains unchanged:
“Someone will have to clean up all that AI-generated code.”
Sometimes that is true. Sometimes the same is true of code delivered by a human developer.
The useful question is whether the system delivers a satisfactory result, how much supervision it requires, and what the complete process costs.
Hallucinations and mistakes still happen. They deserve verification. They do not justify dismissing the entire capability.
AI does not have to be perfect to change the economics of development. It has to be useful enough, often enough, at an attractive total cost.
Testing has changed the calculation
A capable coding agent can generate and execute hundreds of tests. It can investigate a failure, change the implementation, and rerun the relevant checks.
For a human team, extensive testing competes with deadlines, budgets, and other priorities. Many projects never receive the testing their developers would like to provide.
AI can make more of that work affordable.
Of course, test quantity is not proof of quality. Two hundred tests can repeat the same mistaken assumption. A system can pass every automated check and still calculate the wrong business result.
Someone must understand the requirements and judge whether the evidence supports the conclusion.
But arguing that AI code will inevitably be untested and unreliable ignores how these tools can actually be used.
What happens to the cheap-coding business model?
Companies have spent years moving development offshore to save money.
Suppose an offshore developer costs 30% of an onshore developer’s rate. That looks attractive when both deliver comparable work.
Now suppose an AI coding agent can complete a particular task faster, at a substantially lower cost, and with comparable or better results.
What happens to the original calculation?
The offshore developer is still cheaper than the onshore developer. But that comparison no longer settles the purchasing decision.
Customers can compare both options with an AI-assisted delivery process.
This puts pressure on anyone whose main proposition is inexpensive implementation capacity. It also challenges expensive developers who cannot explain the additional value they provide.
Geography does not determine competence. There are excellent engineers in lower-cost markets and poor engineers in expensive ones. The vulnerable business model is selling coding hours primarily on price.
If implementation becomes cheaper, you need another reason for customers to choose you.
Replacing experience with cheaper AI operators is a dangerous bet
Some companies appear to see a simple opportunity: remove expensive developers, hire cheaper replacements, and expect those replacements to steer the AI.
That can work when the new team has the necessary expertise. But if it lacks experience, an AI subscription does not automatically fill the gap.
The departing employees may have understood why a strange business rule exists, which source system can be trusted, and why an apparently obvious architectural improvement would break three downstream processes.
Those contributions rarely appear in a count of completed tickets.
In data engineering, someone must understand the landscape.
Which system owns the customer definition? What does revenue mean across departments? How should late-arriving records affect published reports? Which transformations belong in shared infrastructure? What happens when a pipeline fails halfway through?
AI can help explore and implement answers. The business still needs someone accountable for deciding whether those answers fit its circumstances.
Without that understanding, a team can produce more code while making the overall environment harder to operate.
The financial result may look positive initially. Payroll falls. Development output appears healthy.
The consequences can emerge later through inconsistent metrics, fragile integrations, expensive incidents, and changes that take longer because nobody understands the dependencies.
Experienced experts can suddenly achieve more
My view is that companies need a stronger concentration of experienced experts and less capacity devoted solely to implementing detailed specifications.
AI gives those experts greater reach.
An experienced data engineer can use it to investigate unfamiliar code, prepare migration tooling, implement checks, compare approaches, and automate repetitive work.
An architect can explore alternatives with working prototypes. A consultant can spend more time understanding the customer’s problem because implementation requires fewer hours.
This creates an opportunity to tackle work that previously remained on the backlog.
Smart companies will ask:
“What can our experienced people now accomplish that used to be too expensive or too slow?”
Other companies will focus almost entirely on how many salaries they can remove.
Cost reduction is a legitimate benefit. But removing the expertise needed to direct and evaluate the work can undermine those savings.
A company can become cheaper to staff while becoming more expensive to run.
Freelancers have an advantage: speed
Large organizations often move slowly. New tools may require procurement reviews, security assessments, legal approval, training, and changes to established processes.
Those requirements can be necessary. They also take time.
As a freelancer, you can often evaluate a new capability and adapt your workflow much faster, while respecting your customers’ confidentiality and security requirements.
You can test a better coding tool, improve your verification process, or develop a new service without waiting for an organization-wide rollout.
That flexibility matters.
A project that once required several people may become manageable for one experienced professional. Smaller engagements may become commercially viable. You may be able to serve customers whose budgets previously could not support the work.
Your opportunity is to turn that capacity into a stronger offer.
Stop selling yourself as a coder
“I write SQL and Python” tells a customer which tools you use.
It says little about why they should hire you.
Consider offers such as:
- “I investigate slow Teradata workloads and implement verified improvements.”
- “I find why your reports disagree and correct the underlying data logic.”
- “I assess your migration risks and build a practical execution plan.”
- “I automate a recurring data process and establish checks that make it dependable.”
Each describes a customer problem and a responsibility you can take on.
You still need technical depth. In fact, technical understanding helps you challenge plausible AI output, recognize missing requirements, and design meaningful tests.
But learning another syntax alone is unlikely to be a sufficient response to this market.
Learn more about the business. Understand architecture, operations, costs, and failure modes. Develop the ability to take a problem from the first customer conversation through to a working result.
Then use AI throughout that process where it improves the work.
Take the initiative while the opportunity is open
If you are a freelancer, start with a problem you already understand well.
Use current AI tools on a realistic example. Measure the time involved, inspect the result, and test the assumptions. Learn where the tools help and where your intervention matters.
Build a repeatable service around that knowledge. Create an example you can show prospective customers. Expand your customer base beyond people looking for another pair of hands to write code.
None of this makes the current market easy. People losing work deserve more than a lecture about having the right attitude.
But we still have decisions to make about what comes next.
We can spend our time insisting that customers will eventually discover AI cannot code. Or we can develop the expertise and delivery methods those customers will need as the tools improve.
You can wallow in self-pity, or you can take the initiative.
I would rather use this opportunity to build a business that can deliver more than it could before.
Need help with query performance or platform costs?
Diagnose plans, skew, spool pressure and expensive workloads. Agree a baseline and acceptance criteria before making changes.