@sean_ai_stories: Has anyone realized LLMs still suck at estimating how long tasks actually take? I literally asked mine to build one new feature. It planned everything perfectly, split into sections, then said each would take 5-6 hours. Done in 10-20 minutes. As an ex-PM I know why. They were trained on all that old corporate ticket T-shirt sizing docs. Nobody writes fresh real-world ones anymore, so they still think in hours or days. I normally just divide whatever they say by 10 or 20 and keep going. Sometimes my LLM even tells me this will take forever so let’s do something else. I say no and we finish in less than an hour anyway.

Sean's AI Stories

Open In TikTok:

Region: GB

Friday 01 May 2026 21:36:01 GMT

951

Music

Download

No Watermark .mp4 (8.62MB) No Watermark(HD) .mp4 (5.19MB) Watermark .mp4 (9.15MB) Music .mp3

Comments

Eternity :

I’m starting to use risk of uncertainty or complexity for my estimates you can do shirt sizes on that

2026-05-02 08:38:53

Felipe Gil | Founder :

I noticed that two he Claude was calculating for me an API implementation for 2 weeks. It was just health routes and simple get routes.

2026-05-01 22:30:56

BenM :

Neither do project leads

2026-05-05 09:06:59

Anh Nguyen :

I faced the exact same, I asked Claude / chatGpT to ask "how long will it take to do these" -> they said 2 weeks -> I said "do all immediately" -> took 10 mins max 😂

2026-05-02 20:46:17

i.still.want.to.believe :

That’s a smart observation. Maybe the reason is more psychological, re you achieved your task in 5m rather than 6h as it would take for an actual engineer to do.

2026-05-02 10:49:59

Kieran Hayes :

no it's not wrong, it's training on data that came before AI 🤣

2026-05-04 17:57:54

wellwkd :

In fairness, I find our project sizing was so poor even with poker

2026-05-02 20:10:27

To see more videos from user @sean_ai_stories, please go to the Tikwm homepage.