We're at an inflection point. Agentic AI can now produce code faster than most teams can validate it — and some companies are already using that as justification to skip building junior developer pipelines altogether. In this episode, Mike makes the case that this is one of the most consequential mistakes tech leaders can make right now.

Software engineering was never just typing. It was always about judgment: understanding requirements, reasoning about edge cases, thinking through security implications, mapping code to business context. AI can augment the typing. It cannot yet replace the thinking.

KEY TOPICS

- The "it worked on my machine" problem at AI scale — hallucinated packages, test cases that return true, code that compiles but doesn't function
- The Waterfall → Agile transition as a reminder that articulating requirements has always been the hard part of software
- Why eliminating junior developer pipelines creates a knowledge time bomb
- How AI-generated vulnerabilities are correlated, not random — one found pattern becomes a scannable attack surface across thousands of repositories
- The 14-day median dwell time + 22-second hand-off window from Mandiant M-Trends 2026
- The WordPress plugin backdoor: patience as a supply chain attack vector
- Why "trust" must be a continuous evaluation, not a one-time event
- Judgment as the irreplaceable core of software engineering

SOURCES & ARTICLES REFERENCED

1. Mandiant M-Trends 2026
  Median dwell time: 14 days. Time from initial access hand-off to secondary threat: 22 seconds.
  https://cloud.google.com/blog/topics/threat-intelligence/m-trends-2026
  Coverage: https://complexdiscovery.com/twenty-two-seconds-to-hand-off-inside-mandiants-m-trends-2026-findings/

2. Georgia Tech "Bad Vibes" — AI-Generated Code Vulnerabilities
  43,000+ advisories scanned. 18 confirmed cases H2 2025, 56 in Q1 2026 (35 in March). True count: 400–700.
  https://research.gatech.edu/bad-vibes-ai-generated-code-vulnerable-researchers-warn
  https://scp.cc.gatech.edu/external-news/bad-vibes-ai-generated-code-vulnerable-researchers-warn

3. WordPress "Essential Plugin" Supply Chain Backdoor (April 2026)
  30+ plugins purchased on Flippa, backdoor planted, activated 8 months later.
  https://thenextweb.com/news/wordpress-plugins-backdoor-supply-chain-essential-plugin-flippa-2
  https://www.techrepublic.com/article/news-malicious-wordpress-plugins-backdoor-april-2026/

Questions or topics for a future episode? Reach the show at [email protected].

Mike Schubert is VP of Technology at Unum Group. The views expressed in this episode are Mike's own and do not represent the views of his employer.

Elite performance in engineering organizations isn't about finding rock stars — it's about building the systems that produce them. In this episode, Mike Schubert breaks down two Harvard Business Review research pieces that share a single underlying thesis: team performance is a design problem, not a talent problem. From the three levers that produce organizational excellence (talent, team, routine) to the learning velocity habits that keep teams improving in periods of rapid change, Mike applies these frameworks to the reality of leading large engineering teams in regulated, high-stakes environments.

Then: a sharp pivot. The same principle applies to AI. The "human in the loop" is the foundational promise of responsible AI deployment — but new research from Boston Consulting Group reveals a structural flaw. When pushed back on, LLMs don't reconsider. They argue. Understanding this changes how you build AI governance into your organization.

• Why heroics don't scale — and what does
• The three levers of organizational excellence: talent, team, and routine
• What "moments that matter" actually look like inside an engineering org (pull requests, sprint reviews, postmortems, incident retrospectives)
• How to expose high-potential engineers to decisions 1-3 levels above their current role
• Three habits that build learning velocity: run more experiments, make curiosity contagious, ask what people are stuck on
• Why polished status updates are killing your signal — and what to do instead
• LLMs and rhetorical manipulation: what the BCG research actually found
• "An impenetrable fortress of data and rhetoric" — what that means for your AI governance model
• The human in the loop only works if the human hasn't already been talked out of it

• "How to Turn Individual Talent into Organizational Excellence" — James Fulton & Todd Warner, HBR, March 2026: https://hbr.org/2026/03/how-to-turn-individual-talent-into-organizational-excellence
• "3 Ways to Build a Superteam" — HBR Management Tip, April 13, 2026: https://hbr.org/tip/2026/04/3-ways-to-build-a-superteam
• "LLMs Are Manipulating Users with Rhetorical Tricks" — Tom Stackpole, HBR, March 18, 2026: https://hbr.org/2026/03/llms-are-manipulating-users-with-rhetorical-tricks
• Image: Stephen Curry, Warriors vs. 76ers, January 2, 2025 — NBA.com/Warriors:https://www.nba.com/warriors/news/gameday-recap-20250102

The views expressed in this content are those of Mike Schubert and are not meant to represent those of his employer.

What does it mean to lead through uncertainty when the technology underneath you is moving faster than your governance, your metrics, and your comfort level? Mike Schubert digs into four things: Harvard research on why sharing uncertainty builds trust; Microsoft Agent 365 and the shadow agent problem; IBM Bob 1.0 for the COBOL/RPG world; and Jensen Huang's token consumption metric. Three stories, one thread: agents proliferating faster than governance, faster than measurement, and faster than most leaders' comfort level.

Have feedback or a question you'd like to have answered on the air? Email [email protected]

• 0:00 - Intro & episode overview
• 1:30 - Segment 1: When to Open Up at Work (Leslie John / HBR)
• 7:07 - Segment 2: Microsoft Agent 365 & agent sprawl
• 14:17 - Segment 3: IBM Bob & the ghost of Microsoft Bob
• 16:28 - Segment 4: AI ROI - are we measuring the right thing?
• 21:56 - Outro

• HBR: When to Open Up at Work (Leslie John, Feb 2026)
• Book: Revealing by Leslie John - Riverhead Books, Feb 26 2026

• Microsoft 365 Blog - Powering Frontier Transformation (Mar 9 2026)
• Copilot Studio - 6 Core Capabilities to Scale Agent Adoption
• Microsoft Agent 365 (GA May 1 2026)

• IT Jungle: IBM Gets Bob 1.0 Off The Ground (Mar 2 2026)
• IBM Bob
• Microsoft Bob (1995) - Wikipedia

• All-In Podcast: Jensen Huang at GTC 2026

Most tech podcasts talk about technology. This one talks about what actually makes or breaks a technology leader: people, change, and the gap between what organizations want and what they can absorb. In this opening episode, Mike lays out his case — who he is, what this show is, and why the conversation he wanted to have didn't exist yet.

Takeaways

• The technology is rarely the hard part — leading the people and the change around it is.

• CTO Unfiltered exists because the practitioner perspective has been missing from the conversation.