SWLW #701: Gazing at the misty AI Frontier, From Vibe Coding to Agentic Engineering, and more.
Oren Ellenbogen · email · 2026-05-18 Reader 🖥️
In the agentic era, software leaders win by raising their ceiling — investing in fundamentals, original thinking, and emotionally-attuned leadership — rather than letting AI flatten everyone to the same elevated floor.
practical · mixed — A curated newsletter of links with editorial commentary aimed at software leaders. Each entry is a practical recommendation: read this, share this, try this framing. The curator's voice is brief but argumentative — connecting selections into a thesis about how engineers should adapt to the agentic-AI moment.
Structure
| Part | Role |
|---|---|
| This Week's Favorite: Karpathy 'From Vibe Coding to Agentic Engineering' | Frames the moment — vibe coding is yesterday; the new game is agentic engineering, and even hiring rubrics must adapt to it |
| Culture section: AI-resistant evaluations, the Bruno Method, Software Fundamentals Matter More Than Ever | Core practical evidence — three entries that all argue against shortcutting the human substrate (evaluation rigor, engineering judgment, foundational books) even while embracing agents |
| Peopleware section: Gazing at the Misty AI Frontier (Elad Gil) + The Critical Shift in What Differentiates Great Leaders | Strategic counterpart — where to build (closed loops AI eats first) and how to lead (emotion, ethics, original thinking over data and logic) |
| Inspiring tweets coda | Aphoristic reinforcement — trust gut over rationalization; endure what is endurable |
Key terms
- Agentic Engineering Agentic Engineering
The successor mode to 'vibe coding': a deliberate engineering practice in which the engineer designs, orchestrates, and constrains AI agents to do substantive work, requiring more judgment and architectural skill, not less - Vibe Coding Vibe Coding
Casual, exploratory AI-assisted coding driven by prompts and feel — implicitly framed here as a starting point one graduates from - Closed Loops
Domains where input, action, and feedback are tightly bounded (and thus measurable) — the territories Gil predicts AI will eat first, and a heuristic for choosing what to build - Raising the floor (not the ceiling)
Ellenbogen's framing: AI lifts average performance toward a higher baseline, but does not yet extend the upper bound of what is possible — original thinking remains scarce because models have little repetition to learn it from - AI-resistant Technical Evaluation
An interview or take-home design that admits AI assistance but is structured (longer horizon, complex problem) so that the candidate's own judgment is still required and observable
Propositions
- explicitThe relevant skill for engineers has shifted from vibe coding to agentic engineering, and hiring practices should screen for the latterconf: high
- explicitTake-homes and evaluations should explicitly allow AI use when the problem horizon is long enough that delegation alone does not solve itconf: high
- explicitReading foundational books remains an unfair advantage even (especially) when agents can execute many tasks for youconf: highsupports[FULL WORKSHOP] AI Coding For Real Engineers - Matt Pocock, AI Hero · p20 — Both claim reading foundational pre-AI books remains the highest-leverage advantage precisely when agents can execute the tasks.
- explicitAI raises the performance floor but does not yet set the ceiling — original thinking remains scarce because models have little repetition to learn it fromconf: high
- explicitAI will displace value in 'closed loops' first, so new products and companies should be built around those domainsconf: high
- explicitcorroboratedThe next decade's leaders will be differentiated by emotion, ethics, and connection — not by data and logic, which AI commoditizesconf: highsupports ←The best response to AI slop, infinite advice, and online noise is from Robin Williams · p13 — swlw-701: the next decade's leaders will be differentiated by emotion, ethics and connection — not data and logic, which AI commoditizes. Acunzo makes the same claim in the creator register: what AI commoditizes is knowing/information; what endures is lived, felt experience.
- explicitcorroboratedBefore coding (or prompting), teams should reach a shared design concept — agents amplify whatever direction you give themconf: highsupports ←SWLW #710: Revised rules of Eng Leadership, When AI costs more than the engineer, and more. · p5 — Direction-less amplification implies the human must supply direction — the sibling SWLW issue's 'reach a shared design concept first; agents amplify whatever direction you give them.'
- explicitdevelopedGut judgment that contradicts a rationalized head decision is a reliable signal that should usually be heededconf: mediumextends ←The hidden pattern behind successful products | Mark Pincus (Founder of Zynga) · p1 — Pincus's 'instincts right 95% / ideas wrong 75%' generalizes 'gut judgment that contradicts a rationalized head decision should be heeded' into a quantified split between durable instinct (trust) and the fragile idea-layer built on top of it (distrust).
- explicitHiring performance engineers in the AI era is structurally similar to hiring cyber researchers — long-horizon, hard-to-fake skill assessmentconf: medium
- implicitAgents in a well-designed harness following good principles can carry you a long way operationally but cannot teach you the underlying fundamentals you would learn from booksconf: high
- implicitSoftware leadership is now defined by curation and synthesis — picking which patterns to install in your team — more than by technical authority aloneconf: medium
Arguments
Hiring rubrics must be redesigned around agentic engineering, not around legacy coding skill strength: high
Explicit premises:
- Vibe coding has given way to agentic engineering (Karpathy)
- Anthropic discovered the best strategy on its existing take-home was delegating to Claude Code
- Longer-horizon problems force candidates to demonstrate judgment even when AI is allowed
Implicit premises:
- If the optimal strategy on your evaluation is full AI delegation, the evaluation no longer measures the candidate
- Hiring should test what the job actually rewards, which now includes orchestrating AI well
- Performance engineers and cyber researchers represent a category of role where surface output is easy to fake but judgment is not — and this category is expanding
Inferences:
- Companies that do not redesign evaluations in 2026 will systematically mis-hire — selecting either AI-prompters who lack fundamentals or fundamentalists who can't leverage agents
- Take-home interviews specifically need a horizon long enough that agent-delegation alone is insufficient
Engineers should invest in fundamentals (books, shared design thinking) even as they adopt agents strength: high
Explicit premises:
- An agent harness can encode the rules in foundational books but cannot internalize them for you
- You can use a technique without understanding it
- Shared design concept must precede coding to keep agent output coherent
Implicit premises:
- Comprehension is a different asset from execution capacity — and it becomes more valuable when execution is commoditized
- Agents amplify clarity and confusion symmetrically, so the cost of fuzzy thinking has gone up
- Career durability depends on assets agents cannot copy, and reading is one of them
Inferences:
- The premium on judgment relative to typing speed will increase
- Teams without a deliberate practice of shared design concepts will produce volumes of plausible-but-wrong agent output
Where to build (closed loops) and how to lead (emotion, ethics, originality) are both governed by the same logic: AI flattens the measurable middle, leaving the edges to humans strength: medium
Explicit premises:
- AI will eat closed loops first (Gil)
- AI models raise the floor but not the ceiling
- Original thinking has little repetition for models to learn from
- Leadership in the next decade will be about emotion, ethics, connection
Implicit premises:
- What is most automatable is also what is most rationalizable / measurable / repeated in training data
- Value migrates to whatever resists tokenization — relational, ethical, original, taste-driven work
- Building outside closed loops is harder but is where defensible product value remains
Inferences:
- Founders should pick adjacent-to-closed-loop domains, not core closed loops, if they want defensibility
- Leaders who optimize for legibility and data-driven justification will lose ground to leaders who can stand behind unjustified-but-correct judgments
Narratives
- Anthropic redesigning its take-home interview after realizing the optimal strategy on the existing one was full delegation to Claude Code
→ Concrete proof that AI capability now invalidates legacy evaluation design — and that the redesign requires explicitly permitting AI plus extending the horizon, not banning AI - Karpathy's hiring framing at minute 18:39 of his talk — the type of challenges worth using in the agentic era
→ Treated as the authoritative reference point for the unity claim that hiring must adapt to agentic engineering - SahilBloom's grandfather's advice: the worst decisions come from rationalizing past a clear gut no
→ Aphoristic support for the proposition that taste/judgment outranks legible reasoning — and by extension, that human judgment is the leadership asset to defend against AI commoditization
Questions addressed
- ✅ How should hiring change now that agents can solve significant portions of standard technical interviews?
Allow AI explicitly but lengthen the horizon and complexity of the problem until candidate judgment becomes the binding constraint; model evaluations on roles (performance engineering, cyber research) where output is hard to fake - ✅ What should engineers invest in to stay valuable as agents commoditize execution?
Fundamentals from books, shared design thinking before coding, and orchestration skill — i.e., become an agentic engineer, not a vibe coder - ✅ Where should founders build new products in the agentic era?
Around closed loops (which AI will eat first) — though Gil's heuristic is referenced rather than developed - ✅ What distinguishes great leaders when AI matches or beats expert analysis?
Emotion, ethics, connection, and original thinking — the things AI raises the floor on but cannot yet originate - ⭕ How does a leader distinguish 'agent-amplified competence' from 'agent-masked incompetence' on their own team?
Not addressed — the newsletter celebrates fundamentals and shared design as protections but offers no diagnostic for spotting the difference inside a real team
Gaps
- incompleteThe 'raise the floor but not the ceiling' claim is repeated approvingly but never operationalized. How does a leader, in practice, recognize ceiling-extending original work as opposed to confidently-presented model-mediated work? Without a heuristic, the framing becomes a flattering self-description rather than a usable filter.
Read next: Read Cedric Chin on tacit knowledge or Tyler Cowen on 'what counts as taste in an AI era' for concrete diagnostics; also Karpathy's full talk for his hiring criteria as a partial answer - incompleteTwo of the curator's headline framings — 'AI raises the floor' and 'AI eats closed loops first' — are pulled toward each other but not reconciled. If AI raises the floor on closed-loop work, the implication is that closed-loop expertise becomes commodity labor while open-loop / relational / ethical work becomes the new premium. The newsletter implies this but does not state it, leaving leaders without an explicit instruction.
Read next: Elad Gil's full 'Misty AI Frontier' piece + a careful read of Yue Zhao's leadership argument against Christensen's disruption framing — the synthesis is more useful than either alone - incompleteThe newsletter implicitly equates 'agentic engineering' (a craft skill) with 'agentic leadership' (a hiring/strategy task) but never separates them. Whether the same person should both code with agents and lead teams that code with agents — and whether the screening criteria are the same — is left ambiguous.
Read next: Bruno Fernandez-Ruiz's 'Engineering in the Compiler Age' slides (referenced here, worth a focused extraction) and Karpathy's actual hiring section to see how he separates IC craft from leadership selection - incompleteThe 'Anthropic AI-resistant evaluation' story is treated as a positive exemplar, but the newsletter does not engage with the obvious tension: a long-horizon take-home where AI is permitted requires significant unpaid candidate time, which is a known equity problem. The curator endorses the design pattern without acknowledging its cost.
Read next: The original Anthropic write-up, plus any of the recent compensated-interview / pair-programming-screen literature, for a more honest cost-benefit picture - illogicalThe endorsement of 'gut over head' (SahilBloom tweet) sits uneasily next to the endorsement of 'shared design concept before you code' (Pocock). The first valorizes unjustified judgment; the second valorizes explicit-and-shared rationalization before action. The newsletter treats both as obviously correct without noting that they pull in opposite directions for a leader deciding which to lean on in a given situation.
Read next: Daniel Kahneman on when intuition is trustworthy (closed environments with rapid feedback — i.e., precisely the closed loops AI is eating); Gary Klein's recognition-primed decision research for the conditions under which gut beats analysis
Related documents
- 11) AI-spesial med Vilde Opsal og Anders Sveen
- Notes from FOSE Europe
- Eight years of wanting, three months of building with AI
- Podcast: Formal Methods for Every Engineer in an AI-Powered Future
- How AI will change software engineering – with Martin Fowler
- AI Is Changing Software Engineering. The Fundamentals Still Matter.
- [FULL WORKSHOP] AI Coding For Real Engineers - Matt Pocock, AI Hero
- Presentation: AI-First Software Delivery: Balancing Innovation with Proven Practices
- AI Is Changing the Rules of Entrepreneurship
- All models, no matter how smart, will eventually build systems...