The StartingUp Summer. We're using the slower summer pace to take stock: a look back at a particularly rich year of innovation, a map of what we discovered, and a precise, honest account of what we actually tested, adopted... or dropped. First stop: our drafts drawer, and all the content AI made obsolete.

Our back office keeps a record of 58 article drafts written between 2020 and 2022, never published. We reread all of them this summer. Nine of them asked real questions, the kind that deserved a full article: how to evaluate a developer job offer, which commands to memorize to save time in the terminal, when to give in to a site builder. We will never publish them.

Not that they were bad. Their questions simply stopped being asked. In three years, generative AI has done to our drafts drawer what the smartphone did to folding city maps: it didn't improve it, it made it pointless. This article is the honest autopsy of these nine drafts, and of what it teaches us about what still holds value in 2026.

Which content has AI made obsolete? Four families above all: step-by-step tutorials, collections of tips to memorize, tooling debates, and generic career advice. And three things gained value in return: lived experience, a stance you actually own, and the question no one is asking yet.

Nine drafts, nine dissolved questions

An amusing detail we noticed while rereading them: seven of these nine drafts were unfinished. One stops a sentence short of delivering its solution. Another promises twelve tips and only writes three. Even back then, something in us probably knew that the effort of writing the answer would outlast the question's shelf life.

The draft (2020-2022)The question at the timeWhat made it pointless
What questions to ask in an interview?Evaluating an offer beyond the posted salaryA company-specific analysis grid now generates in thirty seconds
12 command-line tipsCollecting commands that save timeYou describe the intent in plain language, the agent writes (or runs) the exact command
Can TailwindCSS coexist with WordPress?Making two front-end toolchains work togetherThe build plumbing configures itself; nobody looks for that tutorial anymore
Adopting the right developer mindsetLearning to search, daring to ask questionsAI answers every "dumb question", with no judgment and no queue
Premium theme or site builder?Trading off delivery speed against clean codeCustom code now ships at builder speed, without the debt
Starting a Next.js + Tailwind + SCSS projectThe clean Monday-morning boilerplateThe tutorial was outdated within six months; the agent generates an up-to-date project in one command
Getting rid of @wanadoo.fr addressesLinking a domain and a mailbox step by stepThe exact DNS records come from pasting your case to the AI
Stopping WordPress from reinstalling its themesA minor irritant, a config constantOne question, one line of answer: the 500-word article no longer has a reason to exist
10 films to (re)ignite your urge to start a companyThe recommendation listicleAI generates one on demand, tailored to the reader's taste

Four families of content made obsolete by AI

Step-by-step tutorials

Installing Tailwind in a WordPress theme, initializing a Next.js project, setting your MX records at your registrar: this content lived by its precision, and died by its precision. Our Next.js draft cited Node 16 and Tailwind v2; it was already outdated before it was even finished. Today, an agent reads today's documentation, not the documentation from its publication date. A frozen tutorial can't win that race.

Tips worth collecting

"I bet you don't know all of these!", promised the title of our draft about the command line. It was right, and that no longer matters at all. It did hold some nice finds, like this alias some of us still use:

But the tip collection rested on a premise that's now gone: memorization as a competitive edge. We no longer store commands, we describe the intent. Encyclopedic knowledge of the terminal has moved from a skill to a curiosity, the way mental arithmetic did after the calculator.

Tooling debates

Should you accept Elementor to move fast, at the cost of a heavy site and real technical debt? Our 2022 draft argued the point with conviction. The debate has dissolved from the middle: custom code now ships at a speed comparable to a builder's. When both options are fast, all that's left is the question of quality, and that one isn't up for debate.

Career advice

The most unsettling family. Our draft aimed at junior developers taught them two things: using a search engine well, and not fearing dumb questions. It contained this sentence, written completely straight-faced in 2022:

That's why they pay a real human with a brain, someone who can investigate and learn.
StartingUp draft, 2022, never published

Four years later, the sentence hasn't become false: it has become the question. What "a real human with a brain" is worth when the investigating is delegated to agents is precisely what every team, every leadership, is now renegotiating.

What hasn't moved (and is worth more than before)

If AI carried away these nine drafts, it left intact, and even more valuable, three things we find everywhere in our practice over the past year.

  • Lived experience. The account of what actually happened on a project, dead ends and real costs included, can't be generated. It's the raw material of everything we write here.
  • A stance you own. AI lays out the options better than any article from 2022. Choosing, committing, and signing your choice remains a human act, and it's the one that inspires trust.
  • The well-asked question. Our nine drafts answered questions everyone was already asking. Value has shifted toward the questions no one is asking yet.

What this shift has changed in our own working methods would deserve its own account, numbers included: here, we deliberately stick to what it teaches us about content itself.

Rereading your own drawer

Our nine drafts are nothing special. Every company has its own drawer of inherited questions: an FAQ written in 2021, a training path designed before assistants existed, a content plan answering searches nobody phrases that way anymore. Rereading them with this lens is an uncomfortable exercise, and a remarkably profitable one.

Because the disappearance of a question is never a pure loss: it signals that a problem has become trivial, and that the attention it used to demand can move to the ones that aren't yet. In three years, our drawer hasn't emptied out, it has changed nature. Obsolete questions have made room for better ones: that is, as far as we know, the most honest definition of progress.