Whether AI is a threat or a lever for your career comes down to two questions: how much of your work is routine production, and how much AI multiplies the judgement you bring. Below is an interactive map of roles, from an architect who works with AI agents on production codebases every day.
Before you read on: where does your role sit?
- A. Amplified: AI makes me faster and I lose little.
- B. Reinvent: AI is taking half of what I do, and I need to move toward the other half.
- C. At risk: most of my day is routine work AI already does well.
- D. Steady: my work is physical or in person, and AI barely touches it yet.
Drop your answer in the comments. I share where I think engineering is heading at the end.
The pattern I keep seeing
I spend most of my working week inside production codebases for real-time trading platforms: market data streams, trade blotters, risk and P&L sign-off, and the SDKs that several delivery teams build on. For the last year I have done that work with AI agents beside me every day, not as an experiment but as the normal way of shipping.
One pattern keeps repeating, and it is not the one the headlines describe. AI does not replace jobs in one go. It takes the routine production work inside a job first: the boilerplate, the first draft, the reconciliation, the test scaffolding, the summary. The same tools then make the people who own judgement, systems and outcomes far more productive.
So the useful question is not “will AI take my job?” It is “which half of my job do I spend my time in, and which way am I moving?”
Two lines decide whether AI is a risk or a lever
- Routine work AI can take. Work that is repetitive, rules-based and easy to check against a right answer: drafting, coding to a spec, matching, coding a medical chart, translating a paragraph.
- How much AI multiplies the role. Work where one person with AI can now do what used to take a team: designing systems, judging trade-offs, owning outcomes, validating models, directing others.
Put any role on those two lines and four groups fall out. Tap a role on the map to see how it worked before, what is happening now, where it is going, and the move I would make in that seat. Roles in my own world of engineering and capital markets are marked with a teal dot.
What I see in engineering and capital markets
Architects who stay in the code are amplified. The gap between a design and running code used to be measured in sprints. With agents, I can scaffold the services, contracts and tests for a design in hours, then argue about trade-offs with something real in front of us. The architect who only draws diagrams loses ground. The one who stays close to the codebase gains a team’s worth of leverage.
Software engineers are in the reinvent group. Routine feature work, CRUD services and UI implementation are being automated now. What is left, and what grows, is steering agents across a whole repository, reviewing what they produce, and owning what reaches production. Domain depth matters more than ever: an engineer who understands how a repo trade settles or how a P&L gets signed off will direct an agent far better than one who only knows the framework.
The codebase itself has to change. The teams I see getting the most from AI are the ones that make their systems readable to it: clear boundaries, good contracts, and an SDK that explains itself. I wrote about one approach in shipping AI agents inside the SDK itself. AI is only as good as the context you give it, and context is now an architectural concern.
QA is turning into AI evaluation. AI writes tests quickly, and many of them test the wrong thing. Someone has to design the strategy and judge the output. That is a new discipline, not a shrinking one.
Trade operations and reconciliation are the most exposed. Matching, breaks, confirmations and settlement checks are rules-based and repetitive, which is exactly what AI agents are built for. The people who come through well are the ones who learn to design and tune the automation rather than be replaced by it.
Quants, risk analysts, platform and data engineers are amplified. AI writes the boilerplate and drafts the commentary and the pipelines. Validation, governance, observability and data quality become the job, and they matter more as more of the system is written by machines.
AI does not remove the need for judgement. It removes the work that used to hide how little of our day we spent using it.
Beyond engineering, the same shape
The pattern holds well outside software. Lawyers sit in the reinvent group: research and first drafts move to AI, while advocacy, client trust and liability stay with them. Paralegals and legal researchers, whose day was mostly that first pass, are more exposed. Doctors are amplified: AI drafts the notes and reads the scan first, and the clinician decides. Medical coding, translation and rules-based public casework are at risk because they are repetitive and checkable. Skilled trades and hands-on care are steady for now, until robotics catches up.
In every case the split is the same. The routine production work goes first. Accountability, judgement and relationships are harder to automate, and they become the core of the role.
The move to make, whatever your role
- Move from producing to judging. Drafting, coding and checking are being automated. Reviewing, deciding and signing off are not, yet.
- Own an outcome, not a task. Roles that carry accountability keep a person in the loop even when AI does the work.
- Become the person who directs the AI. Every field needs people who set up, check and correct AI on real work. Domain depth plus AI fluency is the combination that holds its value.
- Go deep in a domain. Generic skills are the easiest to automate. Knowing how a specific business actually works is not.
- Make your systems and your knowledge AI-readable. Whether it is a codebase, a process or a playbook, the people who can give AI the right context get the most out of it.
Risk and opportunity are the same change
Most roles hold both a risk and an opportunity, and they come from the same shift. The routine half of the job is going to AI whether we like it or not. The other half, the part that needs judgement, ownership and context, gets bigger and more valuable.
My answer to the question at the top: for engineering, B, moving toward A. Software engineering is in the reinvent group today. The engineers who learn to direct agents, stay close to the domain and own what runs in production will end up amplified. The ones who wait for the change to settle will find it has already happened.
The map is a judgement, not a forecast. I would genuinely like to hear where you would put your own role, and why.
Leave a Reply