Engineering Innovations podcast

If AI writes the code, what do engineers do?

0:00
50:03
Spol 15 sekunder tilbage
Spol 15 sekunder frem

When AI Writes Faster Than We Can Read: James Davis on MAGE and Making Course Materials Accessible

Host Kristin Malavenda interviews Purdue ECE Professor James Davis about what happens when AI can generate software faster than humans can review it and how this may reshape software engineering. Davis describes building Docable, a beta website that ingests PowerPoint, Word, Excel, PDFs and other formats to produce accessibility-remediated versions aligned with ADA-related DOJ rules and WCAG 2.1 AA, motivated by the high time and cost of manual remediation for STEM course materials. He says AI image understanding can generate detailed alt text and equation descriptions, but scaling required using coding agents (e.g., ChatGPT and Claude) to build a roughly half‑million‑line system in months. To trust agent-written code, Davis introduces Model-based Agentic Software Engineering (MAGE): make obligations explicit as engineering models, then use deterministic alignment checks to validate implementations, treating AI outputs as candidates, not verdicts. He argues this shifts emphasis from syntax-level coding to modeling, validation, and adapting to change, enabling higher-quality software and new small-market tools.

00:00 AI Outpaces Reviewers

00:45 Accessibility Deadline Shock

02:56 WCAG And Real Costs

05:24 ChatGPT As Accessibility Aid

06:45 Scaling Beyond One Class

10:36 Building Docable Fast

12:17 Trusting Agent-Written Code

12:38 Models Over Code Reading

20:41 MAGE Governance And Alignment

24:17 Candidate Not Verdict

25:17 Candidates Not Answers

25:53 What MAGE Means

26:37 Make Tacit Knowledge Explicit

28:22 Teach Checks And Alignment

29:39 Slide Deck Modeling Example

32:34 Human Facing Limits

33:08 Machine-Facing Freedom

35:25 Sweet Spot And Safety Critical

36:28 Docable As Case Study

38:54 Are Engineers Obsolete

42:37 New Markets And Boutique Tools

45:46 Advice For Future Students

49:24 Closing Thanks And Links

Link to Jamie's blog on Medium: https://medium.com/@davisjam

Purdue ECE: http://engineering.purdue.edu/ECE

Facebook: http://facebook.com/purdueece

X: https://x.com/PurdueECE

Instagram: https://instagram.com/purdue.ece

LinkedIn: https://www.linkedin.com/company/purdue-ece

Purdue University's Elmore Family School of Electrical and Computer Engineering, founded in 1888, is one of the nation's largest ECE departments and consistently ranks among the best.

Flere episoder fra "Engineering Innovations"