The AI dooooooom thread

[frunk]

Claude is insanely good. Especially compared to 6 months ago when we still had to babysit it constantly.

The result is that we've been burning through our backlog as well as quarterly road map. Which is both good and bad. Good in that we can do more, bad in that our qa can't keep up.

We've been using Devin, and it is spectacular for code development.  We used Copilot before and it was barely worth the time investment but Devin is a completely different story.

[DGuller]

I feel like I've been the only one going in the opposite direction with AI.  In my old company we had ChatGPT Enterprise with access to pretty strong models.  In my new job we have Copilot.    Microsoft's ability to create fucking useless products is unmatched.  I'm actually doing more by hands now than I did a year ago.

[Crazy_Ivan80]

Claude is insanely good. Especially compared to 6 months ago when we still had to babysit it constantly.

The result is that we've been burning through our backlog as well as quarterly road map. Which is both good and bad. Good in that we can do more, bad in that our qa can't keep up.

We've been using Devin, and it is spectacular for code development.  We used Copilot before and it was barely worth the time investment but Devin is a completely different story.

I'll query my colleague,  he's an Ai enthusiast so he'll know about Devin.

Copilot... yes, but no. Don't need a copilot who'll gladly steer the plane into a mountain. We've cancelled that subscription

[Syt]

We ran a test last week. Someone asked if there's an easy way to create a seating chart for 400 people on tables of 10 if you want to keep teams together as much as possible. I suggested putting the task into ChatGPT 5.4 pro with max thinking (I've been using it for my project work with decent success). After an hour it was still crunching the problem.

The "fast" option had a solution in under one minute that was deemed "excellent" by our colleague.

[HisMajestyBOB]

I feel like I've been the only one going in the opposite direction with AI.  In my old company we had ChatGPT Enterprise with access to pretty strong models.  In my new job we have Copilot.    Microsoft's ability to create fucking useless products is unmatched.  I'm actually doing more by hands now than I did a year ago.

That matches my experience with Copilot enterprise. It's useful as an enhanced search engine to point me in the right direction, but it won't give me anything useful directly.

[crazy canuck]

I feel like I've been the only one going in the opposite direction with AI.  In my old company we had ChatGPT Enterprise with access to pretty strong models.  In my new job we have Copilot.    Microsoft's ability to create fucking useless products is unmatched.  I'm actually doing more by hands now than I did a year ago.

This is exactly the problem, people find a use case for a particular niche function, and the folks who don't know better say, see I told you so, all AI is useful.  By the way, another lawyer is heading to being disbarred in Ontario for using generative AI. This time the lawyer checked to make sure the case names were real, but the quotes from the cases he included in his brief were entirely fabricated. 

[Zanza]

MS Copilot is grounded based on your Outlook, Teams, SharePoint and OneDrive content. That makes it a fairly powerful engine to search and aggregate your conversations and files. But beyond that, especially those things integrated onto Office, it feels utterly useless. I don't understand why it does not have proper skills to actually do something with your office documents.

[Jacob]

I thought this was an interesting read:

A sufficiently detailed spec is code

This post is essentially this comic strip expanded into a full-length post:



For a long time I didn't need a post like the one I'm about to write. If someone brought up the idea of generating code from specifications I'd share the above image with them and that would usually do the trick.

However, agentic coding advocates claim to have found a way to defy gravity and generate code purely from specification documents. Moreover, they've also muddied the waters enough that I believe the above comic strip warrants additional commentary for why their claims are misleading.

In my experience their advocacy is rooted in two common misconceptions:

Misconception 1: specification documents are simpler than the corresponding code

They lean on this misconception when marketing agentic coding to believers who think of agentic coding as the next generation of outsourcing. They dream of engineers being turned into managers who author specification documents which they farm out to a team of agents to do the work, which only works if it's cheaper to specify the work than to do the work.

Misconception 2: specification work must be more thoughtful than coding work

They lean on this misconception when marketing agentic coding to skeptics concerned that agentic coding will produce unmaintainable slop. The argument is that filtering the work through a specification document will improve quality and promote better engineering practices.

I'll break down why I believe those are misconceptions using a concrete example.

Full article here: https://haskellforall.com/2026/03/a-sufficiently-detailed-spec-is-code

[crazy canuck]

There is an organization that now tracks the cases in which a court or tribunal comments on the fact that AI used in the submission presented by the parties has generated fabricated cases or legal principles.

In BC alone it happened in 10 cases LAST WEEK.