The spreadsheet worked. It was also carrying too much of the company.
A specialty commercial manufacturer serving construction relied on three connected pieces: a $50,000 annual subscription for preliminary 3D modeling, separate operational software, and an enormous Excel workbook that tracked orders from quote through production.
The employees knew the workbook well. It contained years of formulas, pricing knowledge, and workflow decisions. But a mission-critical shared file can become fragile. Concurrent use could leave people with read-only access, and the workbook suffered a major failure every few years. Its most recent repair cost approximately $100,000.
He did not wait for a software project. He started building.
Around June 1, the COO began working with an AI coding assistant on his own time. The first goal was narrow: recreate the preliminary modeling capability well enough to avoid another $50,000 renewal.
Within weeks, he could generate simple 3D objects. His early estimates repeatedly landed within roughly 1% to 3% of the company's established spreadsheet results. The prototype was no longer an idea. It had earned the right to become a project.
He developed a timeline with guidance from tekRESCUE, presented it to the CEO, and received approval. A structured weekly engagement began around early to mid-August.
Automate the repetitive work. Keep the human safety rails.
The application is meant to move one order through a connected online process without pretending that safety, commercial approval, or exceptions should be automated away.
- 01Dealer builds the quote
The dealer enters dimensions, project specifications, and required documents.
- 02System generates the result
Defined geometry and pricing rules produce a preliminary quote and simple 3D model immediately.
- 03Employee verifies completeness
A person confirms that the submission contains what the company and outside engineer need.
- 04Dealer commits
The dealer reviews and accepts the manufacturer's quoted price.
- 05Independent engineering approves safety
A third-party engineer certifies the design before it can move to manufacturing.
- 06Manufacturing fulfills the order
The approved record moves through production, completion, shipping, and payment.
AI helped build it. The application does not need AI to do its job.
The production quote is not a language model's guess. The application carries forward material prices, labor assumptions, geometry rules, and formulas the company refined over years.
The calculations are deterministic: identical inputs produce identical outputs. The COO is validating every supported model by entering the same data into the established workbook and the new application, comparing the results, and investigating differences.
The company uses its own data and operational expertise to accept results within a few percentage points of established pricing. That is validation, not a promise that software can never contain a defect.
The COO owns the product. We help the build survive the real world.
His idea, code, and business logic
The COO owns the application from beginning to end. He controls the model structures, formulas, priorities, and development work.
Roadmap, architecture, security, and release discipline
Weekly stand-ups, a shared timeline, private GitHub source control, access planning, backup strategy, architecture review, and separate staging and production paths.
We did not replace the COO as the builder. We helped him become a better-equipped one.
We told the AI to vigorously defend every decision.
The COO and tekRESCUE used two independent AI coding environments to attack the proposed architecture. The tools raised objections. The humans compared them, weighted them, and decided what to accept, simplify, or defer.
The review changed the hosting decision and moved separate staging and production environments forward by several weeks. It also surfaced work that must be completed before anyone outside the company receives access.
AI can identify a legitimate problem and still propose an unnecessarily complicated answer. The value came from making it argue its case, then applying human judgment.
Working software, with the hard rollout work still visible.
Simple models work and early pricing aligns within the company's accepted range.
Weekly stand-ups, roadmap accountability, and production-readiness work begin.
Complete model catalog plus login and permission foundations.
A small group of approved dealers receives access after internal validation.
Platform ready for broader dealer use if security and workflow gates pass.
This is an active build, not a victory lap.
The company expects ordinary employee involvement to fall from several hours to approximately 15 minutes per quote, and infrastructure costs are projected near $100 per month. Those are projections until real users prove them.
The legacy workbook will remain available for up to a year for pricing comparisons, spot checks, unusual cases, and fallback. External dealer access remains gated on authentication, permission, and data-isolation testing.
Five lessons from the build so far.
- 01
Your best internal AI builder may already understand the business better than an outside development team.
- 02
A complicated spreadsheet may contain valuable institutional knowledge. Preserve that knowledge while replacing the fragility.
- 03
Move a promising prototype into source control and a controlled release process before it becomes mission-critical.
- 04
Use AI to challenge architecture, not merely to agree with it.
- 05
Automate repetitive creation while preserving human commercial approval and independent safety review.