Skip to main content

A tale of two prompts, frustration and a simple solution.

Image
By

Ever experienced what Simon Wardley brilliantly describes in his talk on Wardley Mapping? I've lived through it myself.

My journey began with a prompt to solve a problem using a CLI tool. AI suggested a wrapper which seemed promising. A few iterations later, we added a web UI and extra features missing from the original tool.

I could've stopped there, but wanting to validate the system, I asked AI to generate test code. It tried but failed to run on my end. Hours of frustration followed as I pleaded with AI to simplify either the code or tests. It refused, claiming it would be too difficult, and we ended up in a circular conversation going nowhere.

Completely frustrated, I started fresh with a crucial difference. My new prompt: "I want to do XYZ in the simplest manner and also need ABC, EFG capabilities. No code yet—let's explore solutions first."

The result? AI suggested existing open-source tools I could connect to my environment. My head nearly exploded! This solution was dramatically simpler AND offered functionality I hadn't even considered but would inevitably need.

Lesson learned from this experience and Wardley Mapping: Start with situational awareness. Figure out what's already built before putting on your builder hat just for the sake of it.

 



Popular Posts

Logffillingitis

By
I'm not against of leaving a trace log of everything that happens on a project what I'm completely against is filling documents for the sake of filling documents. Some software houses that are on the CMMI trail insist that in order to keep or to re validate their current level they need all their artifacts in order but what is missing from that picture is that sometimes it becomes quite a time waster just filling a 5 page word document or an spreadsheet which is just not adequate for the task needed. Perhaps those artifacts cover required aspects at a high degree but they stop being usable after a while either by being hard to fill on a quick and easy manner by someone with required skills and knowledge or they completely miss the target audience of the artifact. Other possibility is that each artifact needs to be reworked every few days apart to get some kind of report or to get current project status and those tasks are currently done by a human instead of being automated. ...
Read more

Are we truly engineers? or just a bunch of hacks...

By
I've found some things that I simply refuse to work without. Public, Centralized requirements visible to all parties involved. I is ridiculous that we still don't have such repository of information available,  there is not a sane way to assign an identifier to the requirements. Then we go with the 'it is all on Microsoft Office documents' hell which are not kept up to date and which prompts my next entry. Version control. When we arrived here quite a lot of groups were working on windows shared folders... now it is a combination of tools but heck at least there is now version control. Controlled environments and infrastructure. Boy... did I tell you that we are using APIs and tools that are out of support? Continuous deployment. First time here, to assemble a deliverable artifact took 1-2 human days... when it should have been 20 minutes of machine time. And it took 1 week to install said artifact on a previously working environment. And some other things that ...
Read more

Qualifications on IT projects. Random thoughts

By
Projects exceed their estimates both in cost and time. Why? Bad estimation would be an initial thought. If you know your estimates will be off by a wide margin is it possible to minimize the range? Common practice dictates to get better estimates which means get the problem broken down to smaller measurable units, estimate each of them, aggregate results and add a magic number to the total estimate. What if instead of trying to get more accurate estimates we focused on getting more predictable work outcomes? What are the common causes of estimation failure: Difficult problem to solve / Too big problem to solve Problems in comunication Late detection of inconsistencies Underqualified staff Unknown. I'd wager that having underqualified staff is perhaps the most underestimated cause of projects going the way of the dodo. If a problem is too complicated why tackle it with 30 interns and just one senior developer? If it is not complicated but big enough why try to dumb it down a...
Read more

Job interviews

By
So after my sabatic period I started to go to different job interviews (most of them thanks to my fellow colleages whom I can't thank enough) and after most of them I feel a little weird. Everyone tries to get the best people by every means possible but then somethin is quite not right. Maybe they ask wrong questions, ask for too much and are willing to give to little in return or just plain don't know what they want or what they need. Our field is filled with lots of buzzwords and it is obvious that some people manage to get jobs only by putting them on their résumé. Then there are some places where there is a bigger filter and filters out some of the boasters. But still it is a question of what do they really need and what questions are needed to weed out those that do not cover minimal aspects required by the job. Don't get me wrong, it is really hard to identify good developers on an interview. It seems that almost no one knows what to ask in order to get insights abo...
Read more