You build software around a large language model (an LLM, such as GPT or Claude). The model reads a request, looks things up and proposes what to do. Your code decides what is allowed. On a trading desk that could be an assistant that answers from the firm's contracts and shows the clause it used, or an agent that proposes a trade, waits for a second person to approve it and books it once. You are judged on what happens when the model is wrong, a reply is lost or the data is late.
Not the same job: the machine-learning engineer trains forecasting models, and the data scientist studies the data. This guide is about the engineer who builds agents a firm can let near its trading systems. You never train a model.
Why it is in demand right now
The adverts appeared in two years. UK job adverts that mention AI agents, counted over the same six months of each year. Contract: 3 in 2024, 159 in 2025, 982 in 2026. Permanent: 2, then 135, then 1,137.
Trading firms are asking by name. A global energy trading firm wanted a lead "to drive GenAI, Agentic AI, and RAG initiatives across their trading and shipping business" and to take proofs of concept "through to production". One October advert was titled "AI Engineer: GraphRAG Commodity Trading". An asset manager described its team as "agentic-first".
They want it safe, not clever. The words that repeat in this autumn's briefs: "prototype to production", "secure", "governed", "human oversight", "audit trails". A demonstration takes a weekend. Putting one into production is the job.
What it pays (2026)
| Role | Pay |
|---|---|
| Permanent, middle | £95,000 |
| Permanent, top tenth | £142,500+ |
| Contract, middle | £560 a day |
| Contract, top tenth | £838+ a day |
Indicative. UK adverts that mention AI agents, all sectors, six months to November 2026 (IT Jobs Watch: contract adverts read 10 November 2026, permanent adverts read 11 November). There is no published figure for energy trading alone.
What the interview really tests
- The model gave a confident answer and it was wrong. How would you have known?
- Your agent sent an order and the reply never arrived. What does it do next?
- Search over 300 contracts returns a clause that an amendment replaced. Why, and what do you change?
- How do you show this release is no worse than the last one?
Where it goes from here, honestly
The adverts do ask for years. One London investment-bank contract this month wants "5+ years in Software, Platform and AI Engineering roles". The same advert then says what counts for the new part: a "proven track record of delivering Agentic AI solutions in production environments, or demonstratable experience building AI tools inside or outside of work setting". Your engineering years count, and the agent part can be built and shown.
Calling a model is not the skill, and the frameworks will change. What lasts is knowing how to test an answer, limit what an agent may do, put a person in the loop and keep a record. The title may fade into "software engineer". The firms will still need the person who can do this next to a trading book.
The roadmap
Stages 1 and 2 are in the Energy Trading & ETRM Skills library already, as guides and projects. The last two are a new course, Agentic AI Engineering for Energy Trading. It is being written now and opens first to a small founding group.
| Stage | Where |
|---|---|
| 1. Know the desk | in the library now |
| 2. Production Python | in the library now |
| 3. Build the agent | new course |
| 4. Make it safe | new course |
Stage 1: Know the desk
Energy Trading Fundamentals and the Practitioner Track. An agent on a trading desk reads contracts, trades and positions. You cannot check its answer if you do not know what a right answer looks like.
- Physical Foundations
- Day-Ahead and Intraday Power Markets
- Power Purchase Agreements (PPAs)
- Trade Lifecycle and Settlement
What you leave behind: one page that follows a single trade from the trader's request to the settled invoice, naming each check on the way and who makes it.
Stage 2: Production Python
Python for Commodities and the Systems Builder Track. An agent is ordinary software around a model, so it needs the same tests, reruns and records as a nightly batch. In the adverts: Python (38%).
- Writing Professional Python
- Production Python for Trading Systems
- Project 1: Price Curve Simulator
- Project 2: Position & P&L Engine
What you leave behind: a public repository with a position and profit-and-loss engine, and tests that protect the numbers.
Stage 3: Build the agent
The new course, first half. One project throughout: an AI assistant for a small power trading desk with a battery, a wind farm and its contracts. In the adverts: LLMs (47%), RAG (33%), LangGraph (15%), MCP (12%).
- LLM APIs in Python: OpenAI and Claude
- Tool Calling and Function Calling
- RAG, Knowledge Graphs and GraphRAG
- LangChain and LangGraph: Agent Workflows
- MCP: Connect Agents to Business Systems
What you leave behind: an assistant that answers from the desk's contracts with the clause it used, and reads the trading book as a named person with that person's rights.
Stage 4: Make it safe
The second half, and the part the adverts call "secure" and "governed". The agent is allowed to act, and you prove what it can and cannot do. In the adverts: Azure (38%), AWS (31%), CI/CD (30%), observability (19%).
- LLM Evaluation and Agent Testing
- Guardrails for Agents That Take Actions
- Human-in-the-Loop Approval and Audit Trails
- Multi-Agent Systems and A2A
- Prototype to Production on Azure
What you leave behind: a trade that goes from a typed sentence to the book with a limit check, a second person's approval and a record of every step, running in the cloud.
If you already write Python for a desk: you have the half the general AI courses cannot give. Skim Stage 1, read Production Python in Stage 2, and put your time into Stages 3 and 4.
Percentages: share of the 982 UK contract adverts that mention AI agents and also name that skill, six months to November 2026 (IT Jobs Watch).
The first thirty days
You can start today with what is in the library. Ten hours a week, half reading and half building.
| Week | Do | Done means |
|---|---|---|
| 1 | Physical Foundations. Day-Ahead and Intraday Power Markets. | You can say what selling 20 MW for one hour tomorrow commits the desk to. |
| 2 | Power Purchase Agreements. Trade Lifecycle and Settlement. | You can read a PPA and say who carries the price risk and who carries the volume risk. |
| 3 | Writing Professional Python. Project 1, the price curve simulator. | Working code in a public repository, with tests. |
| 4 | Production Python for Trading Systems. Then one model call of your own: a trader's sentence in, a filled form out, checked by code. | A script that turns "sell 20 MW from the battery for 6pm tomorrow" into a form, and refuses a sentence that is not a trade. |
Prove it: what to put in front of a hiring manager
The advert asks for "demonstratable experience" (their spelling). These are things a hiring manager can open.
- The agent on GitHub. A typed request read by a model, checked by code against a limit, approved by a second person, sent and recorded.
- A test set with scores. Requests with the right answers written by hand, the score for each model you tried, and what each run cost.
- One failure, written up. The reply that was lost and the trade that would have been sold twice: what your code does about it, and the test that shows it.
- The record of one trade. Who asked, what the model read, which rule passed, who approved, what was booked.
- It running somewhere. Deployed in the cloud with a pipeline that stops a release when the tests fail.
Take these to an interview and nobody asks whether you have used an LLM. They ask how you built it.
Stages 1 and 2 are in the Energy Trading & ETRM Skills community on Skool today. Stages 3 and 4 are the new course.
The founding cohort of Agentic AI Engineering for Energy Trading, and its priority list.
Download the Full Guide
Get the complete guide in PDF format for offline reading and reference.
The full Fundamentals series and the Practitioner Track are in the Energy Trading & ETRM Skills community on Skool: join on Skool.
Download PDF Guide