Safer than most, if you adapt. AI agents can draft requirements, summarise workshops, map processes from logs and write test cases, which removes much of a business analyst's documentation load. What they cannot do is decide what a business really needs, negotiate between stakeholders, or define what an agent should be allowed to do. That last task is becoming core business analysis work.
What do agents take from business analysis?
In a retail bank, a business analyst might spend weeks documenting a process, interviewing stakeholders, writing user stories and preparing test scenarios for a change to account opening. Agents can do much of the drafting.
| Task | Who does it once agents arrive |
|---|---|
| Summarising interviews and workshops | Agent, from notes or transcripts |
| First drafts of requirements | Agent, from templates and inputs |
| Process maps from system data | Agent, drafting |
| Test cases from requirements | Agent, drafting |
| Deciding what the business needs | Person |
| Resolving conflicts between stakeholders | Person |
| Judging trade-offs: risk, cost, customer impact | Person |
| Signing off requirements | Person, with the business owner |
Why does the analyst's role grow rather than shrink?
Every agent a bank deploys needs a specification: what job it does, which sources it reads, which capabilities it has, how far it may act on each, where it must hold for a person, and what counts as a correct answer. That is business analysis. It is also harder than a normal requirement, because the analyst must think about refusals, escalations and the cases where the agent should do nothing.
For example, specifying an agent that handles address changes means deciding which requests it may complete alone, which it must hold, what evidence counts, and what happens if a phone number and an address change arrive together. A business analyst who understands fraud and customer outcomes writes that well. A generic specification produces an agent that is either useless or dangerous.
Analysts also write the cases used to certify agents. A certification run replays an agent against real cases from the business's own operation and scores each answer. Choosing cases that cover the difficult edges, not just the easy middle, is analysis work.
Which skills make you harder to replace?
- Specifying agents: role, sources, capabilities, the depth each needs and limits.
- Writing certification cases that test refusals, escalation and scope, not only correct answers.
- Understanding provenance: which sources are the system of record and which are only declared.
- Domain risk knowledge: in banking, fraud, vulnerability and conduct rules.
- Reading decision records to see whether the agent behaves as specified.
These skills lead into AI operations analyst and agent builder roles. Becoming an AI operations analyst as a former business analyst sets out one route, and moving from business analyst to AI agent reviewer covers a shorter step for analysts who want hands-on experience first.
What does the change look like?
Take an invented example. A business analyst we will call Hiroshi, at a retail bank, was asked to document requirements for an agent that answers card dispute queries. He realised the important part was not the happy path but the boundaries: when the agent must hold, what it must never say about liability, and which disputes go to a person straight away. He wrote fifty certification cases covering those edges. The agent's first certification run failed two gates, and his cases showed exactly where. He became the bank's lead analyst for agent specifications.
How does this compare with neighbouring roles?
Compliance and fraud analysts face similar shifts: agents take the volume, and people keep the judgement. See whether a compliance analyst job is safe from AI agents and whether a fraud analyst job is safe. For the wider picture, read AI agent jobs and careers.
What should you do now?
Learn how governed agents decide, including depth, the bounds every action must meet and the four verdicts. Ask to work on any agent project in your organisation. If you want to design and certify agents yourself, a builder-level certificate with a capstone agent shows that skill. When you apply, a cover letter for a business analyst applying for AI operations jobs helps frame your experience.
Questions people ask
Will agents write requirements on their own?
They will draft them. Someone still has to decide whether the requirement is right, which needs knowledge of the business and its stakeholders.
Is business analysis the same as building agents?
They overlap. Builders design, train and certify agents; analysts specify what the agent should do and test whether it does. Many analysts move into builder roles.
Do I need to code?
No. Specifying an agent needs clear thinking about the business, its risks and its customers.