24 Comments
User's avatar
A. BEKINSTED's avatar

We automated the worker and accidentally promoted ourselves to management

Ben Tossell's avatar

without an increase in pay 😭

jocusrisus's avatar

always good to read

Ole Jensen's avatar

Loved this format!

Daniel Urke's avatar

This was a great post - loved being able to compare your process to mine on another similar task I was working on. Would love to see more on context windows and how to preserve the learnings between. That has bitten me multiple times before, especially in larger / longer analysis tasks

Ben Tossell's avatar

thanks! yeah for sure, planning that.

Becky Schmidt's avatar

Liked this a lot. I enjoy seeing how people chat with AI and try to harness. Also - would you be willing to share the extension? I would use it! I have to do this all the time, it’s very annoying as you mentioned.

Ben Tossell's avatar

amazing thank you! well i could (maybe i should do this for future posts too?) but equally you could take my prompts and use the good versions to build your own 😊! or give the post to your agent and ask it to build it! i’d be interested in the results

Becky Schmidt's avatar

Good point haha. I will test this out and let you know!

Zoe's avatar

I like this post because I can imagine showing it to someone unfamiliar with Codex/Claude Cowork/etc. to show them the kinds of things you can ask, and how to ask better to do what you want. especially the annotated prompts/responses. definitely more for the n00bs, but they need stuff too :)

Ben Tossell's avatar

thank you, that was the aim 😊

Matthias Michael Hoefler's avatar

It was very practical to see your process and where things went sideways that were preventable.

It was a lot of work putting this post together, though, it looks like!

Ben Tossell's avatar

yeh i found it useful for myself!

Chris's avatar

Same, I learned so much from this. Like this style of post, appreciate it

Ben Tossell's avatar

awesome thanks chris

Sparsh's avatar

Liked it more than the usual kind of posts!

Lucas's avatar

Would there be an opportunity here to feed those learnings into Codex's memory, so next time you post "Do it" or "Build it," it asks you for more info (or confirmation) instead of just doing it?

Ben Tossell's avatar

yeah could be in your instructions - or even a skill that triggers.

Slava Abakumov's avatar

I didn’t actually like this post.

It could have been compacted to just the learning, and then all other useful bits as usual (links and findings).

Ben Tossell's avatar

thanks for being honest 😊

i purposefully didnt just skip to the learning, i like to learn in over-the-shoulder kind of style, but i get your pov.

usual links/findings will remain on tues/thursday!

Loong's avatar

it's good but I think it's will be more helpful if you add more about how you constructed you prompt and why is it efficient

Ben Tossell's avatar

you can see my prompts… they are not constructed in any way and not efficient 😂

that’s why I say ‘what i should’ve done’. the aim of this is not to show off a perfect process of prompts, because thats rarely going to happen to anyone. its showing how to look at an agent session and see areas for your own understanding to get better so the next session can be better than the last.