Machine Learning Jobs and AI Automation in the UK (2026): What AutoML and Agentic AI Mean for ML Careers

The tooling that machine learning engineers use to build models is increasingly building models on its own. AutoML platforms tune hyperparameters and select architectures with little human input, and a new wave of agentic AI systems can now plan, run and debug parts of an ML pipeline end to end. That naturally raises an uncomfortable question for anyone in, or moving towards, machine learning jobs: if the machine can do the modelling, what is left for the engineer? The honest answer, drawn from UK labour data and current hiring patterns, is more reassuring than the headlines suggest, but it does demand that practitioners move up the value chain.

The Short Answer

Machine learning jobs in the UK are being restructured by AutoML and agentic AI rather than eliminated. The IMF estimates around 70% of UK jobs have high AI exposure, yet PwC's 2025 Global AI Jobs Barometer found roles requiring AI skills grew 7.5% even as total postings fell 11.3%, and carry an average global wage premium of about 56%. ITJobsWatch put the median UK machine learning engineer salary at roughly £85,000 to £95,000 in the six months to May 2025, up around 26% year on year, with vacancies rising sharply. AutoML automates tuning, feature selection and boilerplate code; demand is shifting towards MLOps, AI governance, evaluation and systems architecture. Employers from Google DeepMind to Revolut and Monzo continue to hire. The roles that shrink are narrow, manual ones; the roles that grow oversee, audit and deploy automated systems.

Which ML tasks is AI automating in 2026?

AutoML has matured well beyond the early "press a button, get a model" demonstrations. Modern platforms automate hyperparameter optimisation, model selection, basic feature engineering and large parts of pipeline assembly. The 2026 development that changes the picture is agentic AI: systems that do not just suggest a model but plan a sequence of steps, write the code, run experiments, read the errors and iterate. Recent research, including work such as the Operand Quant single-agent architecture and various AutoML coding-agent frameworks published through late 2025, points to agents that can ingest documentation, infer sensible fine-tuning parameters and deploy tailored models with limited supervision.

In practical terms, the tasks most exposed are the repetitive, well-defined ones: writing boilerplate training scripts, grid-searching hyperparameters, assembling standard data pipelines, and producing first-draft exploratory analysis. Several commentators tracking the 2026 AutoML landscape argue that a practitioner's worth is no longer tied to writing boilerplate code, but to their ability to architect, govern and audit complex AI pipelines. That is a meaningful shift in what a machine learning job actually involves day to day, even where the job title stays the same.

Will AutoML replace machine learning engineers?

On the available evidence, AutoML is unlikely to replace machine learning engineers wholesale, though it will probably thin out the most junior, narrowly task-based work. The pattern matches the wider labour data. The IMF's 2025 analysis suggests roughly 70% of UK workers are in occupations containing tasks AI could perform or enhance, but it splits that exposure: around 35% high exposure with high complementarity (AI helps the worker), 32% high exposure with low complementarity, and 33% low exposure. "Exposed" is not the same as "about to disappear" — it means a role contains tasks AI can affect.

For ML engineers specifically, complementarity tends to be high. Someone who understands the maths, the data and the business problem can supervise an agent, catch its mistakes and make judgement calls it cannot. The growing concern across the field is the "black box" problem: when an agent builds a model, tracing why it made a decision is hard, and if you cannot explain it, you cannot govern it. Explainable AI has shifted from a nice-to-have to a hard requirement, and that work needs people. The likely outcome is fewer engineers doing manual tuning and more doing oversight, evaluation and architecture.

What does the UK automation data actually say?

UK studies broadly agree that AI reshapes work unevenly rather than triggering a uniform collapse. The Institute for Public Policy Research (IPPR) estimated that up to 8 million UK jobs could be at risk in a worst-case scenario, warning that 11% of tasks workers currently perform are exposed, potentially rising to 59% in a later wave as the technology handles more complex processes. Crucially, IPPR framed this as a design choice, not destiny, and called for a job-centric industrial strategy. The most exposed tasks it identified — routine cognitive work such as database management and scheduling — are precisely the kind that sit at the edge of ML work, not at its core.

PwC's 2025 Global AI Jobs Barometer, drawn from close to a billion job adverts, offers the more optimistic counterweight. Productivity growth in the industries most exposed to AI rose from 7% (2018–2022) to 27% (2018–2024). Jobs requiring AI skills grew 7.5% year on year while total postings fell 11.3%, and they command a substantial wage premium. The reading that fits both reports: machine learning jobs sit on the augmented side of the divide, where the technology raises output and pay rather than removing the role.

Which ML roles are growing in the UK?

Demand is concentrating in roles that sit above the modelling layer. As AutoML and agents handle more of the build, the scarce skills become deployment, reliability, governance and the ability to translate business problems into well-specified ML systems. ITJobsWatch data illustrates the underlying heat: machine learning engineer vacancies showed a strong year-on-year rank improvement into May 2025, with the count of permanent roles requiring the skill rising sharply over the period.

The table below sketches the broad direction of travel. It is indicative rather than a forecast, and individual employers will differ.

The clearest winner is MLOps. As models proliferate, someone has to deploy them reliably, monitor drift, manage retraining and keep production systems honest — work that agentic tools assist but rarely own end to end. Salary signals back this up: MLOps engineer pay in the UK was reported around a median of £85,000 in 2025, with London ranges stretching towards roughly £93,000 at the upper quartile, according to market salary trackers.

Which roles are most exposed, and how should you adapt?

The roles most exposed are narrow ones: a junior whose value is mainly writing standard scripts, running grid searches and producing routine notebooks. These are the tasks AutoML and agents do well and cheaply. That does not make early-career entry impossible, but it does raise the bar — entry-level workers were among the groups IPPR flagged as more exposed across the economy generally.

Adapting is less about learning a single new tool and more about climbing the stack. Three moves stand out. First, build genuine MLOps and deployment depth, because production reliability remains stubbornly human. Second, develop governance and evaluation skills — interpretability, bias testing, model assurance — areas the Department for Science, Innovation and Technology (DSIT) and the Alan Turing Institute have pushed up the agenda through the AI Skills for Business Competency Framework, published on 8 December 2025. Third, sharpen the judgement that agents lack: framing the right problem, choosing the right trade-offs, and knowing when an automated answer is wrong. Practitioners who can supervise and audit AI systems are far better positioned than those competing with them on raw model-building.

Who is hiring ML talent in the UK, and where?

Hiring remains broad and is not confined to a single sector or city. Google DeepMind continues to anchor world-leading research from its London base, spanning reinforcement learning, generative models and applied work in healthcare and energy. Fintech is a major employer: Revolut recruits heavily across London, with reported average ML engineer pay around £79,989, while Monzo applies machine learning to customer-experience prediction and search, with senior data roles reported up to roughly £140,000. Defence and industrial players such as BAE Systems also recruit ML and AI talent, broadening the geographic and sector spread.

Beyond the best-known names, growing employers include Quantexa in fraud and decision intelligence, Peak in decision-intelligence software, and Luminance in legal technology. Geographically, demand is heaviest in Greater London — DSIT-funded research put it at around 2.2% of all job postings — but Northern Ireland, the South-East and clusters such as Cambridge and Edinburgh also feature, and remote and hybrid roles widen access further. With the UK economy showing what one analysis described as a sizeable pool of unfilled data-specialist roles, the structural shortage of skilled people has not gone away despite the rise of automation.

How is the day-to-day work of an ML engineer changing?

The shape of the job is shifting from author to orchestrator. Where an engineer once spent days hand-tuning a model, that engineer may now brief an agent, review its plan, inspect its outputs, and spend the saved time on harder questions: is the data representative, is the objective right, does the model behave acceptably in edge cases, and can its decisions be explained to a regulator or a customer? This is more demanding judgement work, not less.

It also raises the premium on collaboration and communication. As ML output becomes cheaper to produce, the bottleneck moves to deciding what to build and proving it is safe to ship. That is partly why product-facing and governance-facing skills increasingly appear alongside technical requirements in UK job adverts, and why the wage premium for genuine AI skills has held up. Far from deskilling the profession, AutoML and agentic AI appear to be raising the floor of what counts as routine and pushing the definition of valuable work upward.

Frequently Asked Questions: Machine Learning Jobs and AI

Will AutoML make machine learning engineers obsolete?

It is unlikely to make them obsolete, though it will probably reduce demand for narrow, manual tasks such as hyperparameter tuning and boilerplate coding. The IMF estimates around 70% of UK jobs have high AI exposure, but much of that exposure is complementary, meaning AI assists rather than replaces. ML engineers who move into MLOps, governance, evaluation or architecture are well placed.

Are machine learning salaries in the UK still rising?

Recent data suggests yes. ITJobsWatch reported median UK machine learning engineer salaries in the region of £85,000 to £95,000 in the six months to May 2025, with a year-on-year median increase of roughly 26%. PwC's 2025 Barometer also found that roles requiring AI skills carry a meaningful wage premium, so demand for genuine skill has held pay firm.

Which ML roles are safest from automation?

Roles that supervise, deploy and govern automated systems look most resilient. MLOps engineers, ML systems architects, model evaluation and assurance specialists, and AI governance roles all involve judgement, reliability and accountability that current agentic tools assist with but rarely own outright. Applied scientists working on genuinely novel problems also remain in demand across UK employers.

What is agentic AI and how does it differ from AutoML?

AutoML automates specific steps such as tuning and model selection. Agentic AI goes further: it plans a sequence of tasks, writes code, runs experiments, reads errors and iterates with limited human input. In ML work, that means agents can attempt larger portions of a pipeline end to end, which is why oversight, evaluation and governance skills are becoming central to the job.

Do I still need strong maths and coding for ML jobs?

Yes. The foundations matter more, not less, because you will increasingly need to check, debug and challenge what automated systems produce. Understanding the underlying mathematics, data quality and model behaviour is what allows you to catch an agent's mistakes and make defensible decisions. UK employers and bodies such as the Alan Turing Institute continue to emphasise these core competencies.

Where in the UK are most ML jobs based?

London dominates, with DSIT-funded research putting AI skills demand there at around 2.2% of all job postings, ahead of Northern Ireland and the South-East. Strong clusters also exist in Cambridge, Edinburgh and other research-heavy cities, and remote and hybrid arrangements broaden access across the country, so candidates are not strictly tied to the capital.

How can early-career candidates compete with AutoML tools?

Focus on what tools cannot easily replicate: problem framing, deployment, governance and judgement. Building MLOps and evaluation skills, demonstrating an understanding of explainability, and shipping real projects end to end all help. IPPR flagged entry-level work as more exposed economy-wide, so showing you can supervise and improve automated systems, rather than merely produce routine output, is the strongest differentiator.

Summary: AutoML, Agentic AI and the Future of UK ML Careers

AutoML and agentic AI are automating the repetitive core of model-building — tuning, boilerplate code and standard pipelines — while leaving the judgement-heavy work firmly in human hands. UK evidence points to reshaping rather than wholesale replacement: high AI exposure across the workforce, but with strong complementarity, rising salaries and a clear wage premium for genuine AI skills. The roles growing fastest sit above the modelling layer in MLOps, governance, evaluation and architecture, and major employers from Google DeepMind to Revolut and Monzo keep hiring. For practitioners, the path forward is to climb the stack, supervise the tools rather than compete with them, and build the deployment and assurance skills the market increasingly rewards.

Ready to find your next role? Explore current opportunities at machinelearningjobs.co.uk.