Every engineering leader is being asked the same question right now: where does AI fit in how we build things? Most of the answers focus on the output side: AI that writes code, drafts test cases, summarizes defects, and generates documentation.
There’s a quieter question that matters more: what is the AI working from?
In engineering, the answer should be your requirements. And if those requirements are vague, duplicated, out of date, or scattered across Word files and spreadsheets, AI won’t fix the problem. It will amplify it, at machine speed.
AI makes ambiguity expensive
A human developer who gets an unclear requirement usually does something helpful: they walk down the hall and ask. An AI agent doesn’t. It takes the most plausible reading and turns it into hundreds of lines of code, a full test suite, or a polished design summary in seconds. The result looks confident and complete, even when it’s built on a misunderstanding.
That changes the economics of requirements work. Rework has always been the most expensive way to find a requirements problem. AI shortens the time between “vague requirement” and “lots of work built on it,” so each defect in the source travels further before anyone notices.
Three things get worse when requirements are weak:
- Quality drifts. Ambiguous words like “fast,” “user-friendly,” or “as appropriate” can’t be verified by a person or a model.
- Traceability breaks. AI-generated code and tests with no link back to an approved requirement are an audit finding waiting to happen.
- Truth gets fuzzy. If the AI is pulling from a stale document, it will faithfully produce the wrong thing. Which version of the requirement is the current, approved one?
What AI actually needs from requirements management
Strip away the hype and the discipline is the same one engineers have practiced for decades. AI just raises the stakes on getting it right:
- Well-formed requirements. Clear, atomic, testable statements written to a recognized standard such as the INCOSE Guide for Writing Requirements.
- A single source of truth. One governed place where the approved requirement lives, with attributes, status, and ownership.
- Links to everything else. Requirements connected to design, code, work items, and tests, so that people and tools can follow the thread in either direction.
- Controlled change. Versions, baselines, and a clear record of who changed what and why.
- Humans who decide. AI can recommend. People approve, especially where safety, compliance, or contracts are involved.
Notice that none of these are AI features. They are the foundation AI features depend on. That’s where IBM Engineering Requirements Management DOORS Next earns its place in the conversation.
Where DOORS Next fits
DOORS Next has always been about governing requirements at scale: complex products, long lifecycles, regulated industries, and supply chains where a requirement may cross organizational lines. Those same strengths map directly onto what AI needs.
1. Quality at the source
The cheapest place to catch a bad requirement is while it’s being written. IBM’s requirements quality capability scores requirements against industry standards and gives feedback and suggested wording improvements. It began as the Watson-based Requirements Quality Assistant, trained around INCOSE writing rules, and IBM now offers requirements quality analysis through its IBM Engineering AI Hub add-on, as task-level AI agents that sit inside DOORS Next. In practice that means ambiguity gets flagged before it’s baselined and before an AI or a human builds on it.
2. A digital thread that AI can follow
DOORS Next shares data through OSLC, ReqIF, and open APIs, linking requirements to related engineering artifacts across the IBM Engineering Lifecycle Management (ELM) suite, such as work items in Workflow Management and test cases in Test Management. That connected structure is what lets you answer the questions AI raises: Which requirement does this generated test verify? What breaks if this requirement changes? Traceability is the point where requirements management stops being documentation and becomes engineering infrastructure.
3. Configuration management: which truth are we talking about?
AI is only as reliable as the version of the requirement it’s handed. DOORS Next uses streams, baselines, and global configurations to manage versions and variants, so a team (or a tool) can work against a specific, approved state of the requirements rather than whatever file happened to be on a shared drive. For product-line and multi-release programs, that’s the difference between “the AI used our requirements” and “the AI used the right requirements.”
4. AI that assists instead of replacing judgment
IBM’s own framing is worth noting. It has said that in domains where accuracy and compliance are non-negotiable, generative AI can’t be applied naively, and it positions the AI Hub agents as recommending while people decide. That’s the right posture. The AI Hub’s first use cases include requirements quality analysis, natural-language interaction with requirements (answers, summaries, translations), and helping engineers get up to speed on long-running work items. IBM has also published a Requirements Intelligence Assistant demonstrator, built on watsonx.ai, for asking questions about a DOORS Next module, along with candid notes about its limits, such as model context windows and the fact that requirement text is sent to a cloud service. Those are exactly the questions your team should be asking of any AI tool.
5. Governance and auditability
If you work under standards like ASPICE, ISO 26262, or DO-178C, “the AI did it” is not an acceptable answer. DOORS Next provides the controlled change history, access control, and review workflow that make AI-assisted work defensible. You can show who approved a requirement, what changed, and what it links to.
A fair-minded note on positioning
Nearly every requirements management vendor now advertises AI features, so the AI label itself no longer differentiates much. The better question for any evaluation is what the AI sits on top of. A conversational assistant wrapped around a loose pile of documents is a very different thing from AI working over governed, versioned, linked requirements.
That’s the honest case for DOORS Next: not that it has AI, but that it provides the governed foundation that makes AI trustworthy. Competing platforms are moving the same direction, with approved requirements flowing into developer tools and AI agents, so the advantage goes to organizations that have done the groundwork, whichever tool they use.
A few cautions we give our clients:
- Check what’s licensed and deployed. AI Hub is an add-on to ELM, not something every DOORS Next installation has switched on. Confirm version, entitlements, and hosting before planning around it.
- Treat data exposure as a design decision. Know where requirement text goes when an AI service processes it, and choose which projects are appropriate.
- Keep a human in the approval loop. Use AI scores and suggestions to speed review, not to skip it.
Getting your requirements AI-ready
You don’t need a big-bang transformation. A practical starting list:
- Consolidate. Get active requirements out of scattered Word and Excel files and into a governed DOORS Next project. ReqIF, CSV, and document import make this more tractable than it sounds.
- Clean up the language. Run quality analysis on your highest-risk modules first and fix the vague, compound, and untestable statements.
- Define your link model. Agree on link types and traceability expectations, such as requirement to work item to test, so the thread is consistent.
- Baseline deliberately. Establish when and how requirements become “approved,” and make sure tools and people reference baselined content.
- Pilot AI on a bounded problem. Start with quality scoring or summarization on one module, measure the result, then expand.
- Stay current. The AI capabilities arrive through ELM releases (7.2 shipped in December 2025), so upgrade planning is part of the strategy.
The bottom line
AI doesn’t make requirements management obsolete. It makes it the highest-leverage discipline in the building. Teams with clear, traceable, well-governed requirements will get real, defensible value from AI. Teams without them will get faster confusion.
If you’re running DOORS Next, you may already own much of the foundation. If you’re planning an ELM upgrade, migrating from DOORS 9 or documents, or trying to figure out how AI fits without compromising compliance, Strongback can help with assessment, upgrade and configuration, and team enablement.
Ready to get your requirements AI-ready? Contact us at strongback.us.
Sources
- IBM Engineering Requirements Management (product page)
- IBM to launch “task-level” AI agents for engineers (Techfinitive)
- Try it yourself: Requirements AI assistant for DOORS Next (IBM Community, Daniel Moul)
- DOORS Next solution brief (IBM)
- Jazz Community Blog: DOORS Next category (ELM 7.2, December 2025)
- Product Requirements Management for Software Startups (Jama Software)
- Jama Connect MCP: Features in Five