This video is a channel update rather than a regular tutorial. Arjan says up front that it contains no code and lets viewers who came only for a lesson leave early. The agenda has three parts: personal moves, the direction of the channel, and how AI is changing software work. The tone is honest and reflective rather than instructional.
The house he bought with his family turned into a major renovation story. The building dates from the early 1900s and served in turn as a residence, a hotel, a psychiatric ward and an office. Registering it as a home again and arranging the mortgage took months of effort. Four months of demolition followed: three layers of ceilings came down with century-old power and phone wiring plus decades-old network infrastructure. Contractor work blew past even a pessimistic budget, so every item is now ranked by what must be fixed now and what can wait.
The second move is work-related. For half a year he has worked in an old prison building in the center of Utrecht, a location he loves with walks toward the Dom Tower. But there is no air conditioning, windows cannot be opened, and the building will soon close for a large renovation. The new office is also central, close to the station and air conditioned, though a plain standard room. After some decoration the background of his videos will change again.
Across five years of the channel the stable core has been Python plus software design. As Python moved beyond small scripts into machine learning pipelines, data systems, APIs and web apps, content about managing that complexity found its audience. A broad crowd without a classic computer science background started shipping real systems in Python. Teaching design to that crowd fitted the channel and felt natural to him.
He taught programming and design for years at university, where Python was only a small part and most courses were C++ and Java. During his research period he built full applications combining graphics engines with scripting systems. The channel also hosted testing, cloud and hardware reviews covering laptops and keyboards. Still, the core never changed: software design first, Python second.
Today he spends most of his time designing rather than writing code by hand. Apart from video examples, the active project is business tooling: invoice and accounting automation plus a web portal for bulk corporate license sales. He defines the domain, draws the architecture and plans the roadmap for a product with deep Stripe integration and a REST API. The code itself is written by a developer who relies heavily on an AI coding assistant, named in the video as Claude Code.
He sees AI as a major disruption whose impact is largest in software. Systems now find and fix bugs, write tests and build fresh features from scratch. Tooling changes within weeks, so the feeling of keeping up is gone. He openly shares his own constant sense of lagging behind.
His sharpest criticism is the clash between design values and generative defaults. Good design simplifies, deletes code and narrows concepts to the need at hand, while generative systems multiply output by default. That mismatch produces giant pull requests and fleets of agents shipping while their owners sleep. The missing question is one of purpose: why is all of this output needed at all. Without an answer the tool stops being a helper and becomes a token-burning production line piling up software nobody uses.
Comments on the channel split in two: one camp asks for agent orchestration, harness engineering and product reviews, while the other is glad the channel stayed clear of hype. He says reviewing tools does not excite him; he made keyboard and laptop videos, yet his passion is design. A pilot analogy sums up the verdict: a pilot never solves physics equations mid-flight but must still understand lift, drag and thrust. Fundamentals in software work the same way: tools change, design knowledge stays. The channel will therefore teach patterns, principles and senior-level thinking instead of touring each new tool release.
The cost of the choice is stated openly: without hype-driven videos growth may slow and views may fall. The request for support is concrete: likes, comments and subscriptions. The second decision ends sponsored segments entirely; even a monthly slot picked from roughly one in a thousand inbound offers cost time and alignment. The business model narrows to a single product, the Software Design Mastery program replacing all earlier courses. It covers core design, system design and design trade-offs, with certificates, a community, extra material and live sessions built around joint design exercises.
AI commentary
"I read this decision as focus rather than retreat. Tools change every week while design knowledge keeps its value, and I believe the channel becomes more useful by staying on that line."
AI assessment
The strongest version of the opposing case is that agents genuinely help when set up well. Multi-agent review setups and human-approved flows speed up tedious checks and free senior engineers for the design calls that matter. Read that way, his stance is not contempt for tools but a demand that tools serve design judgment. The criticism targets output without purpose rather than the instrument itself.
The missing side is the cost of that output at scale. Open source maintainers are swamped by machine-made pull requests, and the Godot episode shows how that load burns out volunteer labor. Security research keeps finding verification gaps in generated code while the review burden lands on senior staff. The video notes the token bill in passing but never prices it: compute spend, triage labor and false confidence stay off the ledger.
On verifiability, two points stand out. The first is interest: a single-product course business is not a neutral verdict but the choice of a creator earning a living from this content. The second is evidence: one assistant anecdote plus general impressions are thin ground for judgments about every tool. Figures such as a one-in-a-thousand sponsorship acceptance rate, a monthly slot cadence and five years of channel history rest on the narrator statement alone.
My practical takeaway is split by audience. Teams shipping from scratch and developers from non-classic backgrounds gain most from a return to fundamentals. Teams chasing fast prototypes across shifting stacks may get more short-term value from mastering agent setups. I read this video as a sequencing proposal rather than an anti-tool stance: design judgment first, tool mastery second.
Sources
10 links; no other published story cites them. Stories sharing a link do not confirm each other; a source's origin is not inferred from how often it is cited.
- @youtube.com ArjanCodes — episode video
- @arjancodes.com https://arjancodes.com/mastery/
- @arjancodes.com https://arjancodes.com/courses/
- @northeasttimes.com https://northeasttimes.com/2026/08/16/why-software-engineers-say-ai-makes-old-school-fundamentals-matter-more/
- @cacm.acm.org https://cacm.acm.org/opinion/redefining-the-software-engineering-profession-for-ai/
- @deeplearning.ai https://www.deeplearning.ai/the-batch/the-ai-engineering-skills-map-in-detail-software-engineering-fundamentals
- @thenewstack.io https://thenewstack.io/ai-generated-code-crisis/
- @theregister.com https://www.theregister.com/software/2026/02/18/godot-maintainers-struggle-with-demoralizing-ai-slop-prs/4206219
- @survey.stackoverflow.co https://survey.stackoverflow.co/2025/ai
- @code.claude.com https://code.claude.com/docs/quickstart
arjancodes · software design · python · artificial intelligence · coding agents