Agentic Engineering Is Just Product Management: The Karpathy vs Gothelf Fight Over Your Pay Band

When Andrej Karpathy coined “agentic engineering” in early 2024, he described a loop where an LLM‑driven agent reads a task list, writes code, runs tests, iterates, and ships — all with minimal human steering. Weeks later, Jeff Gothelf looked at the exact same task list and said, “That’s just product management.”

Both are right. Both are describing a closed‑loop workflow: define outcome → act → measure → adjust. The difference isn’t technical; it’s organizational — and it shows up in your job title, your leveling matrix, and ultimately your pay band.

#The Shared Task List

Step	Karpathy’s Agentic Loop	Gothelf’s PM Loop
1.	Ingest goal & constraints	Discover user problem & business goal
2.	Plan sub‑tasks (chain‑of‑thought)	Prioritize backlog & write specs
3.	Execute (code, config, infra)	Coordinate design, eng, data, legal
4.	Observe results (tests, logs, metrics)	Measure adoption, satisfaction, revenue
5.	Reflect & re‑plan	Iterate roadmap & stakeholder alignment

The table is isomorphic. The verbs differ — “plan” vs. “prioritize,” “execute” vs. “coordinate” — but the control‑theory structure is identical: a feedback controller minimizing the error between desired state and observed state.

#Why the Naming War Matters

#1. Leveling Rubrics Are Title‑Centric

Most tech companies (Google, Meta, Amazon, Stripe) publish career ladders keyed to title families:

• Software Engineer (SWE) → scope = code quality, system design, technical depth.
• Product Manager (PM) → scope = user impact, business metrics, cross‑functional leadership.

If you ship the same agentic loop but your badge says Staff Engineer, you’re evaluated on technical complexity (e.g., “designed a self‑healing deployment pipeline”). If it says Senior PM, you’re evaluated on outcome ownership (e.g., “grew activation 23% via autonomous experiment loops”).

#2. Compensation Bands Diverge

Public data (Levels.fyi, H1B disclosures) shows a 15‑30% median total‑comp gap between L6 SWE and L6 PM at the same company, even when scope overlaps. The gap widens at Staff/Principal vs. Group PM/Director because equity refreshes follow different budget pools.

#3. Hiring Funnels Filter by Keyword

ATS parsers and recruiter Boolean strings still hunt for “agentic engineering” vs. “product discovery.” A resume that frames the work as “built an LLM agent that autonomously refactors microservices” lands in the engineering funnel; “led a cross‑functional squad that shipped self‑optimizing features” lands in the product funnel. Same human, different inbox.

#The Real Skill Set: Outcome‑Oriented Autonomy

Strip away the labels and the core competency is:

Given a high‑level intent, autonomously drive the loop from hypothesis to measurable outcome, recruiting whatever tools (code, SQL, Figma, stakeholder sync) are needed.

This is role‑agnostic. It appears in:

• Platform engineers building self‑serve infra that product teams consume.
• ML engineers closing the train‑eval‑deploy loop without a PM handoff.
• Founding engineers who are the PM for their slice.

#Concrete Mini‑Checklist (Title‑Neutral)

markdown

• Write a one‑sentence outcome statement (metric + target + deadline).
• Decompose into verifiable checkpoints (unit test, canary metric, user interview).
• Automate the observe → decide → act cycle (CI/CD, feature flags, eval harness).
• Publish a weekly delta (what changed, what learned, next bet).
• Negotiate scope vs. latency trade‑offs with stakeholders in < 30 min.

Master the checklist; the title becomes a packaging decision, not a capability gap.

#Strategic Positioning: Choose Your Pay Band Intentionally

Goal	Recommended Framing	Levers to Pull
Maximize base + equity in engineering ladder	“Agentic engineering lead”	Publish internal tech blog, open‑source the agent runtime, mentor SWEs on eval‑driven dev.
Maximize bonus + scope in product ladder	“AI‑native product manager”	Own a north‑star metric, run quarterly business reviews, build a visible stakeholder map.
Stay hybrid / future‑proof	“Technical product owner” / “Staff engineer, product‑focused”	Rotate quarterly: one sprint writing agent code, next sprint defining roadmap.

Negotiation script (use with your manager):

“I’m delivering the same closed‑loop outcome ownership whether you call it agentic engineering or product management. Can we align my leveling to the scope rubric (impact radius, ambiguity handled) rather than the title rubric?”

#Industry Signals

• Google’s “AI‑First PM” role (2023) explicitly requires “ability to drive autonomous experiment loops.”
• Meta’s “Generative AI Engineer” ladder adds “product sense” as a core competency at E6+.
• Y Combinator’s RFS (Request for Startups) now lists “agentic workflows” under both Developer Tools and Vertical SaaS.

These signals confirm the convergence: organizations are rewriting ladders to reward the loop, not the label.

#Conclusion

Karpathy and Gothelf are describing one control loop with two brand names. The market still pays differently for each brand — so own the narrative. Frame your work in the language of the ladder you want to climb, but keep the underlying skill set title‑agnostic. When the next re‑org or promo cycle hits, you’ll have the receipts (metrics, artifacts, stakeholder quotes) to argue for the band that matches your impact, not just your badge.