People need to reasonate

Discussion
1

Introduction and Context

  • I have been a full-time lurker for the last 20 years, using technology for my own benefit and that of my employer or clients.
  • This might be my first forum post about tech, but recent visits and messages from many of you inspired me to write.
  • I want to share my perspective on where your incremental frustration with the product is coming from.

The Nature of Bleeding-Edge Technology

  • The product and technology you work with—using Cursor or its alternatives—is bleeding edge and at the forefront of a technological revolution.
  • It is neither perfect nor battle-tested, yet expectations remain high, especially because you pay for it.

Hype, Learning, and Realistic Expectations

  • Initially, tools like these amaze you, and as you delve deeper, you begin extracting real value. This process is a form of training on how to work with an agentic IDE.
  • As you invest more time and refine your instructions, your frustration grows when things don’t work as expected.
  • Demanding refunds or perfect performance is unrealistic if you truly want this cutting-edge technology to mature.
  • We must learn to cope with limitations like context windows filling up too quickly, and support the mission rather than tearing it down—or switching to an alternative with the same limits.

Yours sincerely,
A frustrated but considerate customer.

1 Like
2

From my experience: our prompts and rules are not at fault they might be able to steer models into some corner, but it’s not how this is supposed to get perfect. You are not the problem but part of the frustration you develop. As your expectations grow to work successfully with an agentic ide, you also have to grow at how you define instructions to the machine and to find a way to maintain context via instructions until there is some hybrid infinite context or attention mechanism.

3

Ive been a technology worker since 1997. Ive built a bunch of ■■■■ that you have touched, has touched you, has spied on you, and others..

Good post..

–

I personally dont take any rage against such machines - we fn built them…

That said - the biggest thing when a monumental shift in the industry is here – GRANULARIZE the problems. Compartmentalize, break it down, delegate, etc…

Our primary failing as a technical group is to trust that we can have a monolithic solution to all our woes.

We cant.

So - what is the best way to fractalize and compartmentalize and granulaize the actions fo systems we know are limited.

We are getting another cycle of this lesson with AI bots, CURSOR, WF, whatever…

Its about abstracting YOUR thinking out of you, organizing it meaningfully, and building a plan.

We are so FN lucky that we now have a tool we can implement incrementaly and we can build UX wrappers to capture, document, and re-use actions, ideas and code tasks in a way that is so much more dynamic.

So my number one recommend for folks is to wrap all agentic actions intoa /scripts/ folder in your project dir, with a table of what they are, proper headers in such files with DESCR, LOGGING, INVOCATIONs, etc…

And a fundamental aspect is the development_diary that is to be continuous smart documentation of the workflow that you explored in building the thing.

The development_diary is in addition to all the other .md/.rmd docs that you build - its a log of the thought process.

Extensively leverage Github CLI access ability of the agent such that youre directing the bots to build GIST and such…

You need to wield the tools via the bot appropriately.

1 Like