
Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
“Technically cracked” still comes from reps, not vibes — the host says the formula for 2026 is the same as ever: build across greenfield, brownfield, and open source codebases, then get human feedback from experienced engineers via places like ADPList.org.
Juniors using AI aren’t doomed — resistant seniors are at bigger risk — he pushes back on the idea that Cursor/Claude-native juniors will stall in five years, arguing that people with 20–30 years of baggage who refuse to adapt are the ones who will become “legacy quite soon.”
Hand-writing code is becoming less central, and that changes what the job feels like — at some organizations, he says, engineers “don’t really write code by hand anymore,” so if your fulfillment comes from flow-state coding and polish, you may need to find meaning in system design, debugging, architecture, or hobbies.
Vibe coding creates the same old software messes, just faster — his acupuncturist Uber driver built a real app with Lovable, analytics, tests, and users, which proves the power of AI tools, but he warns that hard-to-change codebases and security holes are now arriving on a six-month timeline instead of a six-year one.
Hiring is shifting toward system design and adaptability — he’s already seeing companies care less about code trivia and more about whether you can design a system that starts simple, scales, and survives tradeoffs around database and compute complexity.
This is a brutal moment to coast or disappear — whether it’s someone writing only 500 lines of code a year, or a laid-off engineer considering a gap year, his message is the same: the market is “the culling,” and people who don’t keep learning, networking, and adapting will get weeded out.
He opens by saying nothing magical changed in 2026: getting great still means “putting in the reps.” His fitness analogy is the core frame — the best engineers are the ones who’ve built enough that, in a problem discussion, they can already see the step-by-step path while everyone else is still figuring out the holes.
The practical recipe is three-part: personal projects, open source, and mentorship. He especially likes open source because it forces you to code for maintainers instead of for your own taste, and he plugs ADPList.org as a place where engineers from Spotify, Netflix, and elsewhere give 30-minute mentoring sessions for free.
Reacting to the claim that juniors who learn with AI will struggle in five years, he basically says: no, five years is forever in this field. His point is that juniors have an unusual advantage because they start with a blank canvas, while engineers with 20 years of habits may have more trouble swallowing the six-month adaptation period that AI tooling demands.
This section gets personal: he says he spent months not doing hands-on coding and felt the difference. He still loves building, but the old satisfaction of getting into flow, typing out neat solutions, and polishing code with peer review is “completely changed” in some orgs where agents generate most of the code.
He laughs at the Reddit post about a six-month-old codebase that works, makes money, and horrifies any engineer who opens it, then ties it to a real-world story: an Uber driver in Australia, an acupuncturist, used Lovable to build a web product with analytics, tests, and users. That’s incredible to him — domain knowledge plus execution now beats technical polish early on — but he warns that the usual problems of rigidity and refactoring show up much faster, and security is the part that genuinely worries him, especially after the Lovable secrets leak story.
When he reads the post from the engineer who writes maybe 500 lines of code a year and still gets promoted, his reaction is blunt: if you’re okay coasting, fine, but it’s a career risk. He says your skills will atrophy, and if layoffs hit, “four years of experience” won’t mean much if you spent those years scrolling reels and patching ancient apps.
On passive hiring, he says the best engineers often aren’t spamming cold applications because if they already have growth, freedom, or pay, they need something genuinely better to move. His answer is networking: hackathons, prior coworkers, referrals, and fast, respectful interview loops beat generic hiring pipelines and eight-hour take-home tests.
His broader thesis is that software engineering isn’t dying, but the center of gravity is moving. Coding was never 100% of the job anyway — he cites his own YouTube poll of 305 votes where 46% said coding is under 25% of their work — and the durable skills now are system design, architecture, debugging production, infrastructure, business tradeoffs, and communicating how a simple system can scale.
He closes hard: if you just got laid off, now is not the time to disappear for a year. He talks about his own year-and-a-half away from hands-on engineering as a period of major career FOMO, warns that tools are getting more expensive — including GitHub Copilot pricing changes he describes as a jump from roughly $30 toward $300 with multiplier changes — and says this market is “the culling,” where people who can’t learn fast will be weeded out.
Share
Keep Reading
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.

Playbook
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.

Playbook
Learn how tasteful prompting helps you move beyond generic AI output by shaping context, style, and judgment from the start.

Playbook
OpenAI shipped /goal for the Codex CLI. It turns a prompt into a persisted, self-continuing contract.