4–5 November 2026 · SEEC · CU Boulder East Campus
A two-day working meeting on building AI systems that engineers, scientists and the public can rely on — and on being honest about where the line is.
Why we're convening
That gap is a research problem, an engineering problem and a judgment problem at once — and no single discipline owns it. Over two days we are bringing together the people working on all three: computer scientists building and verifying agents, engineers who have to validate AI in the physical world, researchers studying human oversight, cognitive impact and the future of work, people writing the governance that decides what an organization will and will not allow, and industry teams already shipping these systems.
This is a working meeting, not a conference. The goal is a community: shared problems, shared benchmarks, shared honesty about what does not work yet.
See the newest research on making AI trustworthy and responsible, and meet the faculty and students you could partner with to advance your own R&D.
Compare notes with peers across industry on what is actually working, so the hard lessons get shared rather than relearned company by company.
Build the relationships now that let us go after upcoming federal and private funding opportunities together, rather than separately.
Why industry should come
Keynotes from Haifeng Chen (NEC Labs) on trustworthy enterprise software transformation, and Dragos Margineantu (Boeing) on building robust AI-based systems in the wild. Research talks from faculty at CU Boulder, UIUC, UT Austin and Utah cover testing and benchmarking AI agents, verification and guardrails, and AI security at scale.
Hallucinations in specialized domains, verifying AI-written code, governance gaps, and expertise erosion on teams. These are the risks companies are running into now.
Leaders and Practitioners from Amazon AWS, BAE Systems, Boeing, Google, Indeed, Intel, NEC, Workday and others will share what's working, what isn't, and how they're handling it — alongside the researchers who can help solve it.
How the two days work
This community wants both the frontier and the practicalities, and those are different rooms. So we split them across two days, and you are welcome at either or both.
Day 1Wed 4 Nov
For people building, verifying and studying agentic systems.
Day 2Thu 5 Nov
For people who have to use this stuff on Monday.
Topics our speakers cover
Thread
Agent harnesses are software, and software has bugs. Reproducible benchmarks of real harness failures, harness-centric test generation, and agents that diagnose and repair other agents — including hundreds of previously unknown bugs found in production open-source agentic systems.
Thread
What it takes to bound what an agent may do: enforcing temporal and grammatical constraints on generation, hybrid pipelines that pair LLM creativity with static and dynamic program analysis, neurosymbolic reasoning, formal semantics and runtime assurance.
Thread
What an agent is allowed to touch, and how you find out when that goes wrong: reasoning symbolically about the plans agents produce, red-teaming the tools and webpages they rely on, automated software vetting, and keeping multi-step workflows predictable in cost and quality.
Thread
Keeping the human meaningfully in the loop as AI moves off the screen and into the physical world — what makes people trust an AI system in the built environment, what human–robot interaction teaches us about teaming with agents, how agents can adapt to individuals over time, and how to design for people at their most vulnerable.
Panel
Turning responsible-AI principles into policies teams can actually follow — what to write down, what to require, what to audit, and who signs off when an agent is allowed to act on the organization's behalf.
Panel
What sustained collaboration with AI does to human thinking — to judgment, to memory, to the willingness to sit with a hard problem — and what that means for jobs, teams and professional practice over the next decade. Evidence where we have it, honest uncertainty where we do not.
Panel
The role is shifting from using AI to supervising it. What engineers and researchers need to be able to do — verify work they did not produce, hold onto hard-won expertise, and stay effective as the tools move underneath them — and what employers and universities owe them.
Panel
Finance, healthcare, manufacturing and defense have the hardest version of this problem: foundation models saw little of their domain data in training, so the general-purpose recipes underperform — and the constraints are strictest exactly where the models are weakest, with HIPAA and comparable privacy regimes, security review, and data that cannot leave the building.
Tutorial
As AI systems are further oriented towards autonomous workflows and interactions with the physical world, engineers and developers should be equipped to anticipate the sociotechnical impacts of their work. This tutorial leverages core concepts from Responsible AI practice to help participants build their capacity to both imagine and mitigate potential risks as they design and build AI systems.
Panel
What changes in a computing education when students arrive already coding with agents — from people who are teaching those courses now. John Regehr teaches Utah's CS 3960 "Vibe Coding" alongside advanced compilers, bringing a compiler-testing researcher's standards for correctness to a classroom where students generate most of their code. Brad Hayes is teaching a new CU course, "Design, Build, and Think AI." Danny Dig has taught three: Building AI Agents, GenAI-augmented Software Engineering, and AI-driven Leadership. Alongside them, Tushar Udeshi is building AI learning tools at Khan Academy.
Day-two working sessions
Working sessions rather than talks, shaped by our pre-workshop community survey. Five of them are built directly around what industry teams are wrestling with — those are marked below.
For industry teams
Practices for getting verifiable code from AI agents, including how to review code you didn't write.
For industry teams
A working session that ends with a draft AI policy for your team.
For industry teams
A clinic where you arrive with a real challenge and leave with an expert contact and a next step.
For industry teams
A straight assessment of which AI tools are worth adopting, and how to judge new ones six months from now.
For industry teams
How teams can collaborate with AI without outsourcing the thinking.
For industry teams
Panel. What works in finance, healthcare, manufacturing and defense when the foundation model has barely seen your domain — and privacy, security and data residency bound every option.
Where AI belongs in a course, where it destroys the learning objective, and how to teach students to tell those apart — with John Regehr, who teaches Utah's "Vibe Coding" course.
Who is speaking
More speakers from academia and industry are being confirmed now, particularly on the responsible-AI side — governance, cognitive impact, the future of work and durable skills. To be announced soon.
Keynote · Day 1
Head, Data Science and System Security · NEC Laboratories America
Trustworthy Enterprise Software Transformation. Chen leads NEC's data science department in Princeton, where his team works on LLM agents, uncertainty quantification, privacy, and anomaly detection and root-cause analysis in systems ranging from microservices to spacecraft. Over 80 patents and 100 peer-reviewed publications, earning multiple best paper awards.
Keynote · Day 2
Senior Technical Fellow, AI Chief Technologist · Boeing
Robust AI-based Systems in the Wild. What principles for designing and deploying AI systems carry over to new domains, and what has to be answered before those systems are robust at runtime? Dragos works where decisions carry high risk and the data is sparse: the structure of modern AI systems, optimizing for constraints that are not fully specified, acting without a complete model of the real world, and building human-automation systems that are safe. Full biography (PDF)
Attending
Both days are at capacity. Thank you to everyone who registered — we are looking forward to seeing you at SEEC.
If you are registered: check-in and badge pick-up open at 8:00 am in SEEC Room C120, with breakfast. The full programme for both days is on the agenda page, and the Day 1 poster list is published as a PDF.
Anything else — a question, a press enquiry, or interest in where the Center on Trustworthy and Responsible AI goes next — write to Danny Dig at [email protected].
Planning your trip
CU Boulder East Campus. Open in Maps · SEEC floor plan
Denver International Airport (DIA) is the nearest airport, with shuttle, rideshare and rental options to Boulder.
Space in the lots by SEEC is hit or miss, so buy a permit sooner rather than later instead of counting on finding something when you arrive.
The University of Colorado has negotiated reduced rates at a number of hotels in Boulder. If you are travelling in, book from that list rather than at the public rate.
Hotels with negotiated CU rates →
Boulder fills up in the autumn, so book early and mention the University of Colorado rate when you reserve.
Book return flights out of DIA departing after 7:00 pm on 5 November. Day 1 runs through the evening banquet, which is where a good deal of the introductions actually happen — worth staying for.
SEEC — Sustainability, Energy and Environment Complex
Room C120 (the auditorium)
4001 Discovery Drive, Boulder, CO
SEEC floor plan →
Wednesday 4 November 2026 — research track
Thursday 5 November 2026 — practice track
See the detailed agenda →
Organized by the Center on Trustworthy and Responsible AI initiative at CU Boulder.
This workshop is supported by NSF-OAC grant 2138286 and the University of Colorado Boulder.
