menu

Devs my hat is off to you

Backer ●3 ●14 ●48
calendar_today • schedule1 min read

I am not a developer in any way and will never claim to be, I am a systems engineer and that's what I accel at.

Developers and programmers deserve huge respect for all they do!

Over the past few months I have been vibe coding (or plain language coding) and I already had an appreciation for what you do, but after spending hours vibe coding it's tiring.

But I don't vibe code like some that you see, I have a full on conversation with Claude before hand and really flesh out all of the details and foundation, the. Only once I have fleshed everything out I have Claude give me the exact prompt that it would follow to complete what I am doing. Probably the wrong way, but having this convo for a couple hours to dial everything in, works best for me.

If you are in the same boat, how do you approach things?

11 Comments

•

Really enjoyed this. Devs definitely don't get enough credit.

2 votes
•

@[unwind] I thought I knew how difficult it was until I started trying and it's 100x more difficult.

2
•

Well deserved! Developers solve complex problems every day, and their dedication often goes unnoticed. Respect to everyone building and shipping great software

2 votes
•

@[SuMiTa] it really does and gave me more patience to apps I use, except Microsoft; they are worth over. A trillion dollars, they should be able to do things right. Talking about you windows 11, worse than vista and ME together.

2
•

That’s interesting that you have Claude create the exact prompt for you! 😸

I often use Claude in a similar way. First, I ask it to help me create a specification or design plan, then I review and adjust it before moving on to implementation.

Of course, before starting the actual work, I also define agent rules and skills, such as structure, design patterns, coding style, and other project-specific decisions.

It usually helps me get more consistent results, but I still review the output every time because AI can sometimes take a tactical approach just to complete the task, rather than a strategic approach that fits the whole project.

2 votes
•

@[nyaomaru] I also setup a system wide prompt that tells Claude not to just agree or appease me and I also told it that I am not the source of fact and that it is to evaluate its suggestion and my request and notify me if my thinking or process is wrong and to explain why. It got so much better when it stopped telling me I was right and started to be an advisor than and yes item

3
•

@[kkierii] That’s really interesting. I haven’t tried setting a system-wide prompt like that yet, but it makes a lot of sense. I’ll try it out too. Thanks for sharing!

2
•

@[nyaomaru] if you want send me a dm and i can send you what I used as a starter.

2
•

I actually find using multiple different "agents" works better, especially for the planning and spec work. Claude will do Claude things. ChatGPT does it's thing. Gemini does, well, whatever. However, bounce an idea off of one, take that initial spec doc and pass it to the others until everyone, including yourself, says "yeah, that's in good shape." At that point I'll pass it to Claude Code or Cursor to start implementing things.

And it generally goes beyond a single "spec" doc. Things like a folder of Architectural Decision Record (ADR) documents helps tremendously. Include in the planning doc not only what to build, but what is out of scope. Come opinionated with the tech stack you want to use, frameworks, database, everything. Write that into the spec too.

Much like conversations in the UI version of ChatGPT, Gemini, Claude, etc. not being specific about things allows the LLM to infer intent. Similar to saying "I like blue" and getting a Navy Blue UI element when you meant or were hoping for Cobalt Blue. Being specific also tends to reduce token expenditure because the model doesn't have to make those decisions.

Just my experience and two cents here. Happy to discuss further as well as I do this quite a bit. :-)

0 votes
•

@[Ken W. Alger] these are all good points and a lot that I do. When I am close to a decision I'll get a full breakdown of what it thinks is needed and then I'll take that to ChatGPT and see. Theres been times it's went back and forth but eventually come to an agreement.

This is great info and advice. Thanks for sharing.

1
•

@[kkierii] I've discovered that each AI is better at certain tasks, but when used in concert, they do a good job of evening things out and developing a pretty good plan.

1
🔥 Join developers growing publicly
Share your knowledge, build in public, and grow your developer presence with a global community.
Join CoderLegion

More Posts

Your Backup Data Knows More Than You Think. HYCU aiR Is Finally Asking It the Right Questions.

Tom Smithverified - May 14

I’m a Senior Dev and I’ve Forgotten How to Think Without a Prompt

Karol Modelski - Mar 19

MCP Is the USB-C of AI. So Why Are You Plugging Everything In?

Ken W. Algerverified - Jun 10

The Sovereign Vault — A Comprehensive Guide to Protocol-Driven AI

Ken W. Algerverified - Jun 4

Cisco's Amy Chang: A Model's "Passport" Doesn't Tell You Where It Actually Came From

Tom Smithverified - Aug 27
chevron_left
Kerry Kierverified
3.5k Points • 65 Badges
California, USA • blog.vertexops.org
16Posts
27Comments
61Connections
Systems engineer working in public safety, focused on infrastructure that has to stay up when it mat... Show more

More From kkieriiverified

The site is live: The Known Good

Stop Saying Vibe Coding is Easy

Been MIA for a Bit

Related Jobs

View all jobs →

Commenters (This Week)

2 comments
1 comment
1 comment

Contribute meaningful comments to climb the leaderboard and earn badges!