#AI made #coding cheaper, BUT where did the cost move? #agenticengineering #software #vibecoding

Agentic Engineering · 2026-07-05

AI tools like code generators are making the initial creation of software much cheaper and faster. However, this doesn't mean the overall cost of software development is going to zero; instead, the bottleneck has shifted downstream. The new challenge lies in the evaluation, testing, review, and compliance of the AI-generated code, especially in complex, production-grade systems. The focus shifts from generating code to responsibly integrating and ensuring the quality and trustworthiness of the rapid output of AI agents.

What If AI Proves the Riemann Hypothesis and No One Understands It? – @3blue1brown

Dwarkesh Patel · 2026-07-05

This video discusses how AI might contribute to solving complex mathematical problems, like the Riemann Hypothesis. The speaker distinguishes between two types of AI contributions: finding simple, elegant connections between existing mathematical fields (like a 'lightning bolt' between two domains) versus brute-force, long chains of reasoning without new theoretical insights. He suggests that if AI primarily offers novel, understandable theoretical frameworks, it would enhance human comprehension and collaboration, but if it only provides lengthy, opaque proofs, it would be less beneficial for human understanding.

Why Are LLMs Bad at Writing (but Good at Code)? – Grant Sanderson (@3blue1brown )

Dwarkesh Patel · 2026-07-04

This discussion explores the unique contributions of human authors compared to AI in writing, particularly in technical and creative contexts. While AI excel at distilling and explaining existing information, human writing offers valuable elements of the unpredictable and novel, crucial for creating original insights and coherent narratives. The argument suggests that true authorial skill lies in generating new ideas and thoughtfully structuring them, aspects that current autoregressive language models struggle to replicate.

read more

The conversation begins with a question about why the same progress that leads to clean and mergeable code in software development doesn't necessarily translate to clearer writing. The initial counter-argument posits that for much of the writing consumed daily, AI models like LLMs are already superior at explanation and distillation. The speaker notes a personal preference to feed complex human-written text into an LLM and ask it to explain this to me, finding the AI's explanation often clearer than the original human-authored content. This suggests that for tasks focused purely on information transfer and simplification, AI currently holds an advantage.

However, the discussion quickly pivots to distinguish between distillation and explanation of existing ideas, which LLMs excel at, and the generation of new insights and unpredictable, novel content, which remains a core strength of human authorship. When thinking about the quality of an essay or a book, the value isn't just in summarizing existing knowledge or explaining it clearly, but in the original ideas and the way they are structured into a coherent, well-motivated narrative.

The speaker emphasizes that writing, particularly creative or insightful writing, inherently includes an element of the unpredictable. This isn't just about randomly increasing the 'temperature' parameter in an LLM; it's about a deliberate, nuanced choice of when and how to introduce novelty or a surprising perspective that leads to deeper insight. This intentional unpredictability is what makes human-generated content truly engaging and valuable.

The core argument is that the 'book' itself, or any original piece of content, is generated by an author through exploration of ideas in the world, followed by deciding which aspects are interesting and how to present them in a compelling way. This entire generative process, from initial exploration to the final structured narrative, is what constitutes true writing, beyond mere summarization or explanation. If the initial book itself were simply generated by an LLM, the entire chain of value would be different. Therefore, the very act of seeking to understand and distill a human-written book implies a recognition of the original human insight that went into creating it. The elements of exploration, deciding what is interesting, and crafting a coherent, well-motivated narrative are central to what makes human writing valuable and are aspects where current autoregressive LLMs, despite their explanatory prowess, fall short. The human ability to be 'deliberately choosing something that's novel' is seen as directly contradictory to the pattern-recognition and prediction-based nature of current LLM generation.

Loop Engineering

Addy Osmani · 2026-06-07 · 13 min read

TLDR: The next evolution of AI coding workflows isn't better prompting — it's building systems that do the prompting for you. A "loop" is a small automated system that discovers work, dispatches it to parallel agents in isolated worktrees, verifies results against a defined goal, and persists state externally so nothing is lost between runs. Both Claude Code and Codex now ship the five primitives needed (automations, worktrees, skills, connectors, sub-agents) to build these loops without custom tooling — the core insight being that your job shifts from writing prompts to designing the orchestration logic that generates and evaluates them.

The Intent Debt

Addy Osmani · 2026-06-05 · 9 min read

TLDR: Intent debt — the missing written record of why your system is built the way it is — is uniquely dangerous in an AI-assisted workflow because agents can't recover it. Unlike technical or cognitive debt (which agents can refactor or explain away), intent can only be fabricated by a model, not reconstructed. Every agent session you run without externalized rationale multiplies the cost: each agent starts cold, fills gaps with plausible-sounding guesses, and silently drifts your system away from what you actually meant to build.

Run Out to Meet It

Kent Beck · 2026-04-08 · 1 min read

Charity Majors argues that AI has reset everyone's expertise to zero simultaneously, meaning the accumulated advantage senior engineers built through years of mastering clean, readable code is now a liability rather than an asset. This matters because it reframes the career threat from "AI replaces junior devs first" to "engineers who over-invested in craft fundamentals may actually be worse positioned than newcomers with no habits to unlearn."

Your Data Is Made Powerful By Context (so stop destroying it already)

Charity Majors · 2026-03-09 · 10 min read

TLDR: Storing telemetry in separate pillars (logs, metrics, traces) destroys the relational context that makes observability data exponentially more valuable — each additional attribute you store multiplies query power combinatorially, but only if the data stays unified. AI agents are already rejecting three-pillars data in favor of richer, pre-siloed signals because they need that contextual density to validate changes precisely. Joins across silos don't restore what's lost, and at agentic scale this becomes a hard blocker.

My (hypothetical) SRECon26 keynote

Charity Majors · 2026-03-03 · 9 min read

TLDR: Majors has fully reversed her cautious 2025 stance: AI-generated code has crossed a threshold from novelty to fundamental shift, and passive skepticism is now a career liability. SREs are actually well-positioned for this transition — their outcome-driven, guardrail-building mindset matters more than ever — but only if they actively engage rather than waiting for the change to be forced on them.

Why Your Progress Is About The Same As Everyone Else's

Kent Beck · 2026-03-26 · 1 min read

Kent Beck challenges the cognitive bias of benchmarking your progress against a visible peer group, pointing out that you're only seeing a non-representative sample of people who happen to be in the same room. This matters to a senior engineer because it directly affects how accurately you assess your own skill growth, technical debt decisions, and team performance relative to the actual industry distribution.

Interview with Esther Derby – A Leader in Navigating Complexity

Henrik Kniberg · 2024-05-30 · 6 min read

TLDR: Leading in complexity requires a fundamentally different mental model — instead of controlling outputs like a factory, leaders must shape conditions for good outcomes to emerge, like tending a forest. What those conditions look like depends on your organizational level: senior leaders set broad direction and constraints, middle enablers optimize the system, and team-level leaders tend to their immediate working environment. The core shift is learning to lead indirectly — directing through context-setting rather than explicit control — which is conceptually simple but demands a genuinely different way of seeing and acting.