From 931d9692a5583fe968ff44d76517e0fc7c38e2f1 Mon Sep 17 00:00:00 2001 From: David Johnston Date: Thu, 27 Aug 2026 15:18:55 +1000 Subject: [PATCH] Create an_ai_generated_blindfold.mdx --- .../posts/an_ai_generated_blindfold.mdx | 61 +++++++++++++++++++ 1 file changed, 61 insertions(+) create mode 100644 src/routes/posts/an_ai_generated_blindfold.mdx diff --git a/src/routes/posts/an_ai_generated_blindfold.mdx b/src/routes/posts/an_ai_generated_blindfold.mdx new file mode 100644 index 0000000..ce0d6c6 --- /dev/null +++ b/src/routes/posts/an_ai_generated_blindfold.mdx @@ -0,0 +1,61 @@ +--- +meta: + title: An AI-generated blindfold + description: Increasingly, I'm finding it hard to hard to get a sense of human consensus on a topic from the internet + dateCreated: 2026-08-27 + +tags: + - "software_engineering" + - "ai" +--- + +Let's say I've got whatever question, usually about AI; here are real examples of questions I've had + +- How are people testing their coding agent prompts? +- Is there a tool similar to [claude-dev-tools](https://claude-dev.tools/) but for OpenCode? + +Well, I have three main ways of getting an answer: + +- I can ask AI +- I can google it +- I can go onto a web forum and ask. + +I'm finding that for each of these approaches - chances are, I'm getting an AI generated response. + +- Asking AI - is explicitly an AI response +- Googling - I'm going to be seeing the best SEO answers - the SEO being driven by AI +- Going on a web forum and asking - increasingly seems to be AI bots. + +This makes it hard to assess whether people are actually getting value form that answer - or if it's just the convincing expert opinion of an AI. + +Similarly - I can find various tools online, but again it's hard to assess whether anyone is getting value from them, or if it is a vibe coded side project with no real battle testing. + +## It's less of a problem for regular pre-AI software engineering concepts + +Recently I've been doing more with telemetry and observability. + +I use chat-based AI for this and generally I trust the answer I'm getting. + +## It's more a problem for new fields - such as AI itself + +AI is a fast moving field - so it's fair to say 'of course you can't get a clear picture of what the consensus is - there _is_ no consensus'. + +It's a bit like bad old days of JavaScript web frameworks, where every three months or so it felt like there was a new framework, and if you backed the wrong horse you were stuck with a tool that the industry left behind. Like that, but on steroids. + +And there has always been a bit of a problem of cargo cult programming - people adopting a technology without really knowing what problem it is solving. + +But at least back then there were a handful or tools, and they were industry wide. + +Now, the problem is: + +**ease of vibe coding a tool + lack of human connection = the same problem being solved repeatedly by ten thousand different people** + +The AI companies probably love this - rather than having one team use AI to vibe code the tool and the rest of the world use it - have one thousand teams each use AI to vibe the cool and each be using AI to maintain said tool. + +## The solution + +I think in-person meetups are more important than ever. + +Presenting at a meetup 'hey look at this tool I vibe coded, here's how I did it' I think is valuable signal, it gives us a grounded picture of what's actually happening in a fast moving industry. + +