The Extremes Trap: Why AI Deserves More Nuance

A single remark from Daniel Jalkut, a veteran indie developer known for tools like MarsEdit, has travelled further this week than most lengthy analyses of artificial intelligence. He posted it on Mastodon, John Gruber picked it up on Daring Fireball, and Simon Willison then amplified it on his weblog. The remark goes, in direct translation: "My take on AI is, essentially, that those against it are too much against it, and those for it are too much for it".

It is not a quip. It is a rather precise diagnosis of the current state of public debate.

The Problem of Trenches

By now in 2026, the AI debate has been divided for years between two camps that speak to each other little and listen even less. On one extreme are those who see in any AI tool an instrument of labour exploitation, cognitive degradation, or imminent catastrophe. On the other are those who celebrate every model launch as an irreversible step toward universal abundance and dismiss any criticism as thinly veiled Luddism.

The problem with both positions is not that they are entirely false. It is that they are incomplete, and that incompleteness causes them to lose practical utility. A developer who adopts Claude Code with critical thinking, evaluates which tasks to delegate and which not to, and measures the actual result in their workflow, is doing something qualitatively different from both the uncritical evangelist and the sceptic who rejects the tool without having tried it.

Why This Distinction Matters

Nuance is not a comfortable middle-ground position on principle. It is an epistemological stance: recognizing that the effects of AI are heterogeneous, context-dependent, and require case-by-case evaluation.

In the Claude ecosystem, this translates into very concrete terms. Claude Code hooks can automate security reviews in a CI/CD pipeline with measurable results; they can also introduce opaque dependencies that nobody understands when they fail. MCP servers expand model capabilities powerfully; they also add attack surface and operational complexity that many teams are not prepared to manage. Subagents allow delegating complex tasks; and they can also propagate errors silently if supervision is not designed correctly.

None of these realities cancel out the other. They coexist.

Who This Remark Speaks To

Jalkut's observation is particularly useful for three specific profiles.

First, engineering teams that need to justify internally the adoption or rejection of AI tools. Presenting a nuanced evaluation, with use cases where it adds value and cases where it does not, is more sustainable than defending an ideological position that crumbles at the first counterexample.

Second, product and business leaders who face pressure to "integrate AI" without clear criteria for what problem it solves. Jalkut's remark gives them cover to ask the right question: does it help here or not?

Third, those who participate in technical communities, forums, conferences, social media, where AI debate tends to polarize quickly. Citing this position as a starting point is a way to lower the temperature without abandoning rigour.

The Value of Those Who Choose No Side

That the remark has circulated through Jalkut, Gruber, and Willison, three figures with decades of track record in independent software and technical writing, is not accidental. All three have built their credibility precisely by not automatically aligning with any hype. Willison in particular has been one of the most balanced voices in LLM analysis for years: he acknowledges real capabilities, documents real limitations, and sells neither apocalypse nor utopia.

In that sense, his weblog picking up this remark functions as a signal of community validation: there is space, and there is demand, for analysis that does not begin from a trench.

---

We at ElephantPink believe this remark deserves to be taped to the wall of any room where a decision about AI adoption is made. Not because it is neutral, neutral without criteria is worthless, but because it forces you to think before taking a position.