Engineering Management in the AI Era - Dennis Nerush (Talk Outline)

Dennis Nerush — director of AI engineering at Elementor (the #1 website-building platform, 19M+ sites), ~15 years in the industry, 7 practicing AI “before it was cool.” How AI made his team better and worse, and how to redefine the EM role. Four parts: the “impossible job,” where AI helps, the dark side, and the new role.

1. The Price of Over-Reliance

• Opening props: manual gear cars (common in Europe), paper address books (nobody), a Budapest paper map, Stack Overflow — technology makes objects and skills obsolete. In 2006 the US Navy dropped celestial navigation from its syllabus (they have GPS).
• But over-dependence has a price: you remember maybe 1–2 phone numbers by heart (a problem if you lose your phone); some can’t get back to the hotel without GPS; driving a manual after years away is hard. AI is the new revolution with the same trade-off — rely too heavily and you pay a price.
• The EM role is “the impossible job” (Michael Lopp, ex-Slack) for four reasons: never enough time, outnumbered by chaotic beautiful snowflakes, always too much to do/know, and unattainable expectations. It’s a constant balance of managing technology and managing people.

2. Where AI Helps

2.1 Productivity

• ChatGPT before Google (even for a website’s opening hours); note-takers (tl;dv/Fathom-style) that summarize meetings + action items so the manager can be present; Gemini for email summaries/replies in your tone; presentation drafts (Gamma, Napkin AI — turns text into visualizations “I could never draw”); working prototypes for buy-in (Lovable, Base44, Cursor — not production-grade, but far better than words for a conversation).
• MCP connects data sources (Jira, Confluence, Slack) to AI environments (Claude, Cursor): auto-create 15 tickets from a technical design; an incident companion pulling logs, past post-mortems, and codebase to say “I saw an increase here, maybe related to that bug months ago.” (“A new MCP server every minute.“)

2.2 Knowledge Consumption

• ChatGPT scheduled tasks — a recurring prompt (e.g., a daily bite-size summary of The Five Dysfunctions of a Team in your format) delivered as email/push, so learning happens without finding the time.
• ChatGPT Deep Research — scanned 26+ sources (incl. LeadDev) in ~5.5 minutes into a comprehensive report (he used it to prep this talk).
• NotebookLM — turn any text/report into a podcast to listen to while doing dishes, and now an interactive one you can join to ask questions/disagree (released as an app — his go-to podcast player).

2.3 Soft Skills

• Craft measurable performance goals (with enough context); prep weekly one-on-ones from a Notion database of each report’s skills/background and ongoing observations (“based on what you know about David, what should I discuss?” — augmenting, not replacing, his own agenda).
• A custom “Feedback Coach” GPT — asks what/who/context, then teaches a communication model (e.g., SBI — Situation/Behavior/Impact) via your example and outlines how to prepare (a prep technique, not a copy-paste email) so a small piece of feedback doesn’t blow up.
• Voice-mode role-play of hard conversations (a grumpy “Poopics” refusing a no-budget conference request) → come in prepared and confident.
• Hiring: compensating job descriptions (hire for what the team lacks, e.g., project management), better interview questions, candidate scoring, and tailored onboarding (no two people are the same).

2.4 Technology

• Data-flow/diagram generation from the codebase in seconds (for tech design or for a manager to understand the team).
• Agentic editors (Cursor, Windsurf) that do the work (scan, plan, implement — vs. Copilot’s autocomplete), with cursor rules to enforce standards (naming, file structure, planning mode, reuse, logging/observability, typing, tests up front, PR format) and live-updated diagrams/READMEs.
• Devin — the first “fully autonomous AI engineer” living in Slack/Jira/Confluence with its own virtual env (spins up local dev, runs CI, fixes lint, opens PRs) — “like a junior engineer, improving.” Lets EMs raise their head from every PR to be strategic.

3. The Dark Side (Over-Reliance)

• A lawyer lost his license over ChatGPT-hallucinated precedents. A ChatGPT-flavored performance goal (emoji-laden, meaningless) deeply offended the recipient — “your people know you’re using AI; they want to talk to humans.”
• Agentic editors create dependency: “Cursor is down, so we can’t work” — but Cursor isn’t the developer, it’s a tool. AI is great for experienced engineers but dangerous for juniors who vibe-code without understanding, never boy-scout/improve the codebase (→ mess fast), and lose their essence — critical thinking and creative problem-solving.
• Example: told to “write tests” for a bad function, a junior returns 100 tests for code that should have been refactored (single-responsibility violation) — smiling, unaware it’s wrong.
• The “expert beginner” trap accelerates: anyone can build impressive working software in minutes → false confidence that they no longer need to learn best practices/patterns/clean code. Soon we’ll have engineers who’ve never written a line without AI — and it’s on leaders to make them look under the hood and actually grow.

4. The New Role — Balance AI With Being Human

• Beware the dystopian “engineering-manager copilot” (he prototyped one in Lovable) that scans all Slack/PR interaction and auto-sends warnings (“Poopics became passive-aggressive — click to send a warning”). Some love it, some are terrified — that’s the point.
• Instead, remain people-oriented: trust-building, genuinely knowing your people (family names, hobbies, drivers/motivations) and connecting those to the company mission; ensuring people learn and grow and don’t over-depend on AI (use it as a starting point, not the whole solution); giving good words when deserved and confronting when needed. ”Your title makes you a manager; your people make you a leader” (Bill Campbell).
• The new role balances encouraging AI adoption (know the tools, harness the power, give people the chance to learn) with human-centric leadership. Is it harder? “Not harder — different.” Close (Bruce Lee): “Be water, my friend” — adapt to the AI environment while staying focused on growing your team.

5. Q&A

• AI for quality, not just quantity? Depends on context — without guardrails and a definition of “quality” (production-grade, handles load/scale, follows best practices) you get garbage; define them and results improve, but you can’t “fire up AI as magic” — it gets you faster to a point you then tweak.
• Will NotebookLM eliminate conferences? No — like “Video Killed the Radio Star,” radio survived; it’s a different medium for consuming long docs, not a replacement for conferences or podcasts.

People, Companies, Tools & References Cited

• Dennis Nerush — speaker; director of AI engineering, Elementor.
• Michael Lopp (“the impossible job”), Bill Campbell (“your people make you a leader”), Bruce Lee (“be water”), SBI model.
• Tools: ChatGPT (scheduled tasks, Deep Research, voice mode), Gemini, Napkin AI, Gamma, Lovable, Base44, Cursor, Windsurf, Devin, GitHub Copilot, NotebookLM, Notion, MCP, Jira/Confluence/Slack.
• Concepts: over-reliance price, MCP-connected EM workflows, feedback coach, expert beginner, vibe-coding dependency, human-centric leadership.

Video: https://www.youtube.com/watch?v=Xt5LWQmww0M — Transcript via yt-transcript.sh; outline generated from the transcript.