MyBidPilot.us

The Upload That Went Nowhere

By Joey Bess··3 min read·
Wall

When I started MyBidPilot, I knew the idea was too big.

That was not a surprise. The whole thing is meant to help construction suppliers, steel building dealers, and subcontractors handle the messy revenue path: quote comes in, the important details get pulled out, a bill of materials forms, a proposal gets written, follow-ups happen, and eventually a project is created.

In my head, it was one clean line.

Upload quote → extract details → build BOM → create proposal → create project.

In real life, one of those arrows became a brick wall.

The bug that gave me the most grief was painfully simple: the quote upload looked like it worked, but the rest of the system did not reliably move forward.

That is a special kind of annoying bug for a non-technical person like me. If something explodes with a giant red error, at least I know where to point. This one was sneakier. The file would appear to upload. The page would act like things were happening. But the next step — the AI extraction and then the BOM/proposal flow — was not behaving like one connected pipeline.

It felt like mailing a letter, watching the postal worker take it, and then finding out it never left the building.

My first instinct was to ask the AI to “fix it,” which is a sentence I have typed more times than I care to admit. The AI did what AI often does: it came back with a lot of confidence and a lot of suggestions.

Some of the suggestions were useful. Some felt like it wanted to remodel the kitchen because a drawer was stuck.

That has been one of my biggest lessons with this project. The AI can be very helpful, but it does not automatically understand that I am trying to build the smallest working version first. It wants to be thorough. I needed it to be boring.

So I narrowed the problem down.

I told it, in plain terms, that I only cared about the happy path: one uploaded construction quote should turn into extracted information, then a BOM, then a proposal, then a project. No fancy edge cases. No grand architecture lecture. Just follow the ball from step one to step two.

Once I did that, the AI’s advice got more useful.

It basically pointed out that my app had pieces that existed, but they were not all handing the baton to each other in a dependable way. The upload part was one piece. The extraction part was another. The proposal part expected certain information to be available. But “available somewhere” is not the same as “saved in the right place at the right time.”

That sounds obvious now. It was not obvious to me then.

The fix was not glamorous. It was not some brilliant invention. It was making the pipeline more explicit.

After the quote upload, the app needed to clearly trigger the AI extraction. Then the extracted details needed to be saved before the BOM and proposal tried to use them. Then the project creation step had to read from that saved information, not from a wish and a prayer.

I also had to keep reminding the AI not to widen the scope. This project already has enterprise-sized ambition, which is a ridiculous thing for one 50-year-old non-coder to be poking at. If I let every fix become a redesign, I would never finish anything.

When it finally worked as one clean path, I did not jump out of my chair. I mostly just sat there suspiciously, clicking through it again to make sure I had not imagined it.

That is another thing nobody tells you about building with AI: sometimes the victory feels less like triumph and more like relief.

But I’ll take relief.

MyBidPilot still has a long way to go. The full idea is big, probably too big. But that bug taught me something important. In software, especially as a beginner, the magic is often not in making a fancy feature. It is in getting one ordinary thing to move cleanly to the next ordinary thing.

Small takeaway: before making it smarter, make sure it actually passes the baton.