From the Research Universities edition of November 7, 2026
ChemE-as-Code: An Open-Source Platform for Agentic AI in Chemical Engineering
Delft University of Technology (Process Intelligence Research Group, Department of Chemical Engineering) · University research (chemical/process engineering; agentic design platforms) · Netherlands (TU Delft; transferable to research universities and public labs exploring agent-assisted process design, with limits noted by the author)
- Publisher
- arXiv
- Original publication
- 5 November 2026 (arXiv 2611.07947v1); preprint submitted to AIChE Journal, November 2026
- Source retrieved
- 2026-11-08
What happened
Artur M. Schweidtmann (TU Delft Chemical Engineering) introduces ChemE-as-Code, an open-source environment in which LLM agents can carry out chemical engineering work and engineers can check it. A plant is a Git repository of linked, machine-readable files; calculations run outside the LLM; each value records its origin; and automated checks report inconsistencies and out-of-date results by file and line. Skills tell a general-purpose agent how to use tools spanning catalysis, process simulation, plant layout, HAZOP, techno-economic assessment, and life-cycle assessment. In an illustrative natural-gas ammonia case, agents turned a written task into a preliminary basic-engineering package (simulation files, P&IDs, layout, HAZOP, datasheets, relief) for USD 224 of API cost in the design sessions. Planted faults and engineering changes show that the checks find what their rules cover and miss the rest, so they guide human review rather than replace it. No independent engineer has reviewed the design. The author states the case shows the workflow runs and does not establish design quality or a productivity gain, which remains a hypothesis.
Why it matters
Relevant to research universities and public engineering programs that are packaging scientific or design workflows so agents can propose work that humans can still audit. The measured API cost and explicit non-claim on productivity are more useful than vendor speed claims.
Evidence and measured results
Single-author arXiv preprint submitted to AIChE Journal, with an end-to-end ammonia demonstration and a planted-fault/engineering-change evaluation of the checkers. Not yet peer-reviewed; no independent engineering review of the generated design.
Limitations and uncertainty
Preprint; single author; one illustrative ammonia case; USD 224 covers API cost in the design sessions, not total project cost. Checks miss faults outside their rules. No independent engineer reviewed the design; no measured productivity gain. Chemical-process design is adjacent to, not identical with, campus research-lab automation.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-11-07; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Agent-generated process designs can look complete while silently leaving inconsistent, out-of-date numbers across simulations, drawings, safety studies, and inventories.
- Stakeholders
- Engineering school research leadership
- Chemical engineering faculty / PIs
- Research computing
- Environmental health and safety / process-safety reviewers
- Discovery questions
- Are process or experiment files already versioned and machine-readable, or still locked in drawings and PDFs?
- Do automated checks report inconsistencies by file and line?
- Has any independent engineer reviewed agent-produced packages?
- What is in-scope API spend versus claimed productivity?
- Value hypothesis
- A Git-native plant model with out-of-model calculations and origin-tracked values may make agent proposals reviewable at modest API cost (USD 224 in the paper’s design sessions), without establishing design quality or schedule savings.
- Potential engagement
- A single-process pilot that imports an existing design basis into a ChemE-as-Code-style repo and measures what the checks catch versus miss on planted faults.
- Unsupported claims
- Proven productivity gains versus conventional design (author says this remains a hypothesis)
- That passing checks equal independent engineering review (none has been done)
- Safety approval of the illustrative ammonia package
Pre-sales engineering
Role takeaway
- Fit
- Strong where process information can be expressed as linked text or model files; weak where critical documents remain unstructured drawings without digitization.
- Architecture and integration
- Git plant repository plus deterministic calculation tools plus agent skills plus origin/evidence records plus file-and-line consistency checks; a general-purpose agent loads skills per task rather than a large discipline-agent mesh.
- Prerequisites
- Version-controlled design basis and models
- Tool runners for simulation and checks
- Human reviewers for safety and design
- API or local-model access for the agent
- Constraints
- Checks miss faults outside coded rules
- Single illustrative ammonia case
- No independent design review yet
- USD 224 is session API cost only
- Security
- Control who can read plant files and which content is sent to external model APIs; treat HAZOP and cost data as sensitive.
- Proof of value
- End-to-end generation of a declared document set from a written task
- Planted-fault catch/miss inventory by rule class
- Recorded API cost for the design sessions
Delivery
Role takeaway
- Work
- Stand up the repository template; encode check rules; run an illustrative package; plant faults; commission independent review.
- Dependencies
- Faculty process-design ownership
- Simulation tool licenses or open alternatives
- Reviewer time for HAZOP/safety
- Ownership
- Chemical engineering faculty own design acceptance; research computing owns platform operations.
- Skills and adoption
- Engineers learn to review diffs, evidence records, and checker reports instead of only static drawings.
- Governance checkpoints
- No sponsor or instructional use without independent human review
- Checker-miss log reviewed after planted-fault tests
- API egress review for proprietary process data
- Acceptance criteria
- Proposed: declared document set produced and versioned
- Proposed: planted-fault report lists covered vs missed classes
- Proposed: independent reviewer sign-off recorded
- Risks
- Treating green checks as design approval
- Undetected safety inconsistencies outside rule coverage
- Over-generalizing one ammonia illustration
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Represent the plant or experiment as versioned, linked files in Git; execute calculations in deterministic tools outside the LLM; store origin/evidence with each value; and emit file-and-line inconsistency reports that a human can walk.
Governance
Who approves, reviews and stays accountable for outcomes?
Automated consistency checks are a review aid, not a substitute for licensed or independent engineering review. Policy should forbid treating a passing check suite as design approval.
Security and privacy
What data, permissions and controls need testing?
A Git-hosted plant model can contain proprietary process data, safety studies, and cost assumptions. Agent sessions that edit those files need access control, secrets hygiene, and review of what leaves the campus to third-party APIs (the USD 224 figure is API spend). The preprint does not specify a regulated-data architecture.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
May let students and staff inspect origin-tracked values and file-and-line checker reports, but still requires experienced engineers to review what the rules do not cover.
Procurement
What should contracts, pricing and exit terms secure?
If procuring agentic engineering design tools, require out-of-model calculation, origin records, planted-fault evaluations, and independent human review—not a generated drawing demo.
Operating model
Which teams own the service once it runs?
Points toward a faculty-owned design repository with shared platform operations, rather than a fully autonomous design bureau.
Publication history
- 2026-11-07Research Universities · Issue 604 resources
Stable resource ID: cheme-as-code-tudelft-agentic-chemical-engineering-20261105