Blueprint: Engineering in the Age of AI

"If my AI-designed aircraft crashes, it's not Sam Altman or Dario Amodei at the NTSB hearing. It'll be me."

Sesha Manavasi, lead aerodynamicist at FlowBelow, joins Martin Bielicki to talk about where AI is helping the engineers who sign off on physical products, and where it still falls short.

Sesha has worked on both sides of engineering simulation. After supersonics CFD research at the University of Cincinnati, he spent years building and selling simulation tools at CD-adapco, Exa and Dassault Systèmes, then at Luminary Cloud. Now he's the customer, designing aerodynamic devices for Class 8 trucks and trailers.

In this conversation:

• Why release velocity is the wrong goal for engineering software, and the overset mesh release that broke customers' working setups
• Why he thinks AI makes engineering judgement worth more
• The chicken-and-egg problem with physics AI: you spend heavily on validated simulation before it saves you anything
• What a 100x faster solver is really for (exploring more of the design space before launch)
• Where the AI pressure in large engineering companies comes from, and why adoption slows at R&D level
• Whether hardware can ever be tested the way software is
• Why "mechanical engineer" may stop being a useful job title
• How FlowBelow designs, from a 90% failure target in brainstorming to an agentic CFD pipeline

Chapters

00:00 Intro

00:43 From Chennai to supersonic CFD

02:22 Going from tool vendor to tool buyer

04:49 Why focus beats release velocity

08:02 The overset mesh release that burned customers

10:01 A new frontier model every 18 days

11:40 Who's accountable when AI designs hardware

14:30 The physics AI chicken-and-egg problem

17:40 What a 100x speed-up actually buys you

22:11 Organisational inertia

25:06 AI FOMO, and why hardware can't hallucinate

28:50 The engineer's role in five to ten years

30:49 Can hardware be tested like software?

35:39 More mechanical engineers, or fewer?

39:15 Inside FlowBelow's design process

43:47 Where to follow Sesha

Blueprint is Bench's podcast on how AI is changing engineering, through conversations with the founders, researchers and engineers doing the work.

FlowBelow: flowbelow.com

Bench: getbench.ai

Sesha Manavasi: https://www.linkedin.com/in/sesha-manavasi/

Martin Bielicki: https://www.linkedin.com/in/martin-bielicki/

Paul Russell leads the MSc in Integrated Industrial Design at Loughborough University and has taught CAD and design engineering for years. He was early to generative AI, running Midjourney betas in 2021, and now teaches a module on the ethics and aesthetics of generative AI in design practice.

In this episode, Martin and Paul get into:

• Why students entering university this year are being trained for the industry of 2030, not today.
• The AI CAD landscape: kernels vs harnesses, why the real opportunity is a modeller built from scratch, and how PTC, Autodesk and Dassault are responding.
• What happens to CAD teaching when a fully dimensioned three view drawing can be turned into a STEP file in one click.
• Sharing agency with AI inside CAD: from drafting tool back to genuine computer aided design.
• The honest jobs conversation: drudge work gets automated first, and what that means for headcount.
• The skills that stay valuable longest: design for manufacture, material selection and costing.

Blueprint is Bench's podcast on how AI is changing engineering, told through the founders, researchers and engineers working inside it.

In this episode, we sit down with Mark Fuge, Professor of Mechanical Engineering at ETH Zurich and Chair of Artificial Intelligence in Engineering Design, to explore the gap between what industry needs from AI and what academia is actually delivering, and why closing that gap starts with getting both sides in the same room.

Mark shares his journey from writing Perl scripts at GE Aviation to pioneering ML for mechanical engineering back when the idea got you laughed out of job interviews. We discuss his concept of "use-inspired basic research," why the three fundamental challenges for AI in engineering are composition, abstraction, and uncertainty management, and how a helicopter manufacturer's request in 2018 led to an ETH course where students tackle real multi-physics design problems with today's AI tools.

We also dig into his "right turns vs. unprotected lefts" analogy for understanding where AI can reliably take over and where human judgment remains essential, why taste

and design thinking will matter more as building becomes free, and his vision for AI-driven personalised medical devices that could transform care for children with

congenital heart disease.

In this episode, we cover:

• Why AI in engineering is where FEA was in the 1960s, and what that means for adoption timelines
• How engineers are shifting from builders to architects, and why understanding the problem matters more than unlimited compute
• The case for bringing industry and academia together to define the real research questions, not just the interesting ones

Links from the show:

Get in Touch:

Mark Fuge

https://www.linkedin.com/in/markfuge/

Conference Link & Topics

https://event.asme.org/IDETC-CIE

https://idetc.secure-platform.com/a/page/tracks_topics

Martin Bielicki

https://www.linkedin.com/in/martin-bielicki/

Chapters ➡️

00:00 Introduction to AI in Engineering

02:25 Mark Fuge's Journey into AI and Engineering Design 04:03 Bridging the Gap: Industry Needs vs. Academic Research

07:29 Understanding Industry Challenges in Engineering 09:32 Emerging Solutions and Startups in Engineering AI 11:50 Innovative Teaching: Preparing Students for the Future 16:21 The Evolving Role of Engineers in the AI Era

21:56 Perceptions of AI in Engineering

25:50 The Future of Human-AI Collaboration

30:23 Vision for Humanity: Engineering a Better Future

In this episode, we sit down with Steven Holmes, Editor-in-Chief at DEVELOP3D and producer of the DEVELOP3D LIVE conference, to explore what makes AI different from every other technology wave he's covered, and why this time the pressure on engineering teams is real.

Steven shares his perspective from nearly two decades covering product design and engineering software, including why so few teams have moved beyond enterprise ChatGPT licences and what's keeping managers from giving engineers the tools they're asking for. We discuss whether startups or incumbents are better positioned to lead the AI shift, and why legacy software installs are becoming a competitive liability rather than a safety net.

We also dig into what's driving some companies to rethink their entire technology stack, the parallels to earlier industry shifts like cloud CAD, and why the window to act is shorter than most engineering leaders think.

In this episode, we cover:

• Why the gap between AI awareness and actual adoption in engineering is still so wide
• How startups and incumbents are each positioning to win, and where the mergers and acquisitions wave is heading
• What's finally pushing engineering teams to question their legacy tools and move

Links from the show:

https://develop3dlive.com/

Get in Touch:

Stephen Holmes

https://www.linkedin.com/in/stephenholmesd3d/

Martin Bielicki

https://www.linkedin.com/in/martin-bielicki/

Chapters ➡️

In this episode, we sit down with Nomi Yu, a researcher who recently graduated from MIT, where she was co-advised between the DeCo Lab and the Mechanosynthesis group, to explore how AI can enable better parametric CAD generation and why the ethical development of these technologies matters just as much as their technical capability.

Nomi shares insights from her work on GenCAD 3D and the challenge of training AI models when usable CAD data is scarce. We discuss why simply having more data isn't enough, how synthetic datasets can address critical biases, and the potential of federated learning to let companies collaborate on training models without ever sharing proprietary IP.

We also dig into the future of engineering workflows, including why the most successful companies will use AI as a starting point rather than a replacement, and the parallels between "vibe coding" in software and what could become "vibe engineering" in hardware design.

In this episode, we cover:

• Why data quality and bias correction matter more than data quantity for training CAD generation models
• How federated learning could unlock cross-company collaboration without compromising IP
• The case for engineers deepening foundational knowledge rather than racing to automate everything

Links from the show:

https://decode.mit.edu/

Get in touch:

Nomi Yu

https://www.linkedin.com/in/nomiyua6175aadf85/

Raihaan Usman

https://www.linkedin.com/in/raihaan-usman/

Chapters ➡️

00:00 Introduction to the Blueprint Podcast

00:20 Nomi's Journey in AI and Engineering

03:07 Understanding GenCAD and Parametric Design

05:53 Data Quality and Collaboration in AI

10:27 Challenges in Cross-Domain Learning

12:41 Future of Engineering with AI

16:29 Onshape & Their Dataset

19:26 The Future of Engineering AI

26:39 Verification and Trust in AI Systems

34:52 The Future of Engineering Education

43:48 Responsible AI Development and Ethical Considerations

In this episode, we sit down with Abhinav Tanksale, Technical Support Manager at Sentio Technologies and former Senior Crash & Safety Analyst at Magna, to explore the current state of AI adoption in CAE.

Abhinav shares his perspective on where AI is genuinely delivering value today versus where the hype outpaces reality. We discuss Siemens' lead in AI integration, why standardisation at large OEMs can slow adoption, and the practical advice he'd give to managers looking to get started.

In this episode, we cover:

• The state of AI integration across major CAE software platforms
• Why starting with repetitive tasks like geometry cleanup and report writing is the smartest adoption strategy
• The soft skills AI won't replace, and why they matter more than ever

Links from the show:

Abhinav’s Blog

https://myphysicscafe.com/

Get in touch:

Abhinav Tanksale

https://www.linkedin.com/in/abhinav-tanksale-6259b5118/

Martin Bielicki

https://www.linkedin.com/in/martin-bielicki/

Chapters ➡️

00:00 Introduction to Abhinav Tanksale

02:25 The Journey of Abhinav's Blog: My Physics Cafe 04:58 Will AI actually replace CAE Engineers?

07:29 Siemens Digital Thread

08:58 Adoption Patterns of AI in Engineering

11:52 Advice for Managers on AI Integration

13:29 The Limitations of AI in CAE

14:36 The Future Role of CAE Engineers

18:18 Could Standardisation be the Biggest Blocker for AI Adoption in Engineering?

19:38 Envisioning the Future of CAE Workflows

In this episode, we sit down with Moritz, Programme Manager at DeltaVision and former Director of Engineering at Hyperganic, to explore how AI is reshaping engineering workflows.

Moritz shares his perspective on AI's role in breaking down knowledge silos between simulation, design, and manufacturing teams. We discuss the critical balance between AI-assisted speed and the transparency engineers need to confidently sign off on designs, particularly in high-stakes industries like aerospace.

We also dive into Moritz's market research on AI engineering tools, examining the emerging clusters from generative design to physics simulation surrogates, and what it actually takes to get large engineering organisations to adopt new software.

In this episode, we cover:

• Why AI's greatest impact in engineering is democratising knowledge across the value chain
• The four key clusters emerging in the AI engineering software landscape
• What makes engineers actually adopt new tools, and why data consistency remains the biggest pain point

Links from the show:

DeltaVision Hiring:

https://deltavision.space/job-openings/

Get in touch:

Moritz Valentino Leone

https://www.linkedin.com/in/moritz-valentino-leone-b877b41a4/

Martin Bielicki

https://www.linkedin.com/in/martin-bielicki/

Chapters ➡️

00:00 Introduction to AI in Engineering

03:45 AI is best at breaking Silos

07:27 Will AI be the "Final" solution in Engineering?

11:35 The Future of AI in Engineering Design

14:40 Moritz describes the motivation behind starting his blog.

17:20 Emerging Clusters in Agentic Engineering: Simulation 20:30 The Text-to-CAD Cluster

22:35 Adoption Challenges in Large Corporations

27:24 Does having a focussed use case make it easier to adopt software?

29:58 Choose a Workflow to Automate?

31:52 Data consistency in Engineering teams.

"If your job is just CAD modelling and you don't have deeper engineering knowledge, you better start learning - because in two to three years, that skill alone won't be enough."

Ashraf, Design Engineering Manager at Toothsure, shares how his startup has embedded AI into their product development workflow from day one, and why he believes engineers who resist the shift are making a mistake.

In this episode of the Blueprint Podcast, we cover:

• Why privacy concerns are a big blocker to AI adoption in engineering
• The difference between how startups and large companies are approaching AI tools
• How mapping your workflow end-to-end reveals where AI can actually help
• MIT research on AI learning CAD modelling from YouTube videos (link below)

Links from the show:

MIT Research

https://news.mit.edu/2025/new-ai-agent-learns-use-cad-create-3d-objects-sketches-1119

VideoCAD

https://ghadinehme.github.io/videocad.github.io/

Get in touch:

Ashraf Serour

https://www.linkedin.com/in/ashraf-sorour-3953919a

Martin Bielicki

https://www.linkedin.com/in/martin-bielicki/

Chapters ➡

00:00 Introduction

01:04 Identifying Repetitive Tasks for Automation

02:05 Barriers to AI Adoption in Engineering

03:22 How can Data Privacy with AI work in Engineering? 05:41 Exploring AI Tools in Engineering

07:15 Are engineers actually adopting AI?

12:33 Should Engineers focus on Innovation?

14:32 How an Engineering AI Assistant could save time!

17:42 Advice for Engineering Managers to Implement AI 21:32 Conclusion: VideoCAD