One Week, One Idea, Zero Code: How I Vibe‑Coded OpsFixer
What happens if you stop reading about AI‑powered “vibe coding” and actually press build?
From curiosity to capability in 168 hours
A couple of weeks ago I set an experiment: In one week, could a non‑developer like me turn an idea into a risk‑smart app, using nothing but AI tools and spare moments between life’s routines?
The Rise of Chat‑to‑Code
Cursor, Windsurf, and Lovable claim you can chat an app into existence. But marketing pages are cheap; working prototypes are not. My day job sits at the intersection of operations and risk, so I picked a thorny pain‑point I knew intimately: dusty Standard Operating Procedures nobody updates until something breaks or someone gets audited.
Lovable dashboard
The ground rules
Time‑boxed: must be completed within one week, albeit will not be full days solid.
No manual coding: Every line generated by AI.
Security matters: Must include GDPR and role based access controls into the app.
The Journey — Day‑by‑Day Progress & Pitfalls
Day 1 – Spark & Setup Breakfast brainstorming → ChatGPT logo in 90 seconds. Lovable + Supabase scaffolded the UI and database. Challenge: limited uploads to .txt when most SOPs will live in Word or PDF. Learning: prototype quickly, but test against reality on day one.
Day 2 – First Wins, First Regression Process‑map generator produced three stakeholder‑friendly views. Asked AI to “improve layout,” which quietly broke other logics in the background. Learning: every aesthetic tweak deserves a regression test.
OpsFixer - Process Map Summary Page
Day 3 – Security Reality Check Prompted Lovable for GDPR compliance and suddenly needed multi‑tenant controls, row‑level policies, and a working cookie banner. Mistake: bolted security on after UI polish. Learning: build guard‑rails early; the retro‑fit took 30 % of total time.
Day 4 – Risk Engine Online Connected a policy repository to compare SOPs against internal rules and surface variances. Heat‑map view exposed 17 red flags in the demo data. Learning: insight compounds when you centralise metadata once, not per file.
Day 5 – Enhancement Loop Added AI suggestions with accept/dismiss and comment threads. Prompting Lovable to "optimise performance" improved load speed but erased half the buttons. Learning: speed gains are pointless if functionality disappears. Regression test after every optimisation prompt.
OpsFixer Process Overview
Day 6 – Ruthless Declutter Removed duplicate pages, reduced navigation to three clicks: Upload → Review → Improve. Dashboard tiles now surface active risks and accepted improvements at a glance. Learning: "keep things simple stupid"; complexity confuses.
Day 7 – Role‑Based Finish Line Installed granular permissions (Owner, Admin, Editor, Viewer). Quick smoke tests suggest the right people see the right data with deeper penetration testing scheduled for beta. Learning: access control is never truly "done," but it must be "good enough" before user one.
OpsFixer Settings Page
What OpsFixer Delivers Today
Easy SOP ingestion (Word, PDF, TXT).
Multi‑level process maps - exec overview, manager swim‑lane, front‑line checklist.
Policy variance alerts - upload company policies for the SOPs to be compared and ranked against with exportable risk register.
AI improvement engine - suggest, accept, annotate.
Heat‑map dashboard for organisation‑wide view of risk & progress.
Enterprise‑grade security: GDPR controls and granular RBAC.
OpsFixer Risk Engine
Five Takeaways for Leaders
Technology like Lovable makes it extremely easy to build and prototype apps that can help you and your team achieve more. If you're going to try building something for your teams consider the following points:
Start with a real pain‑point. “Cool tech” without a clear use‑case gathers dust. Would this idea help me in the real world? If the answer is yes then try and build it.
Secure first, design second. Audits don’t forgive pretty UIs. Always think about how to make data secure, how data is transmitted, how to keep API keys secure etc etc. The devil is in the detail.
Expect whiplash iterations. Each AI refactor or tweak can fix one bug and spawn two new ones. Test relentlessly and if possible ask others to test.
Humans close the loop. The app can suggest a build, but only the human building it knows if this is what was truly wanted or needed and how people may use it.
Beta early, beta often. Building this is great, and has taught me a lot but opening it up wider will be the real challenge when a more diverse base of real users surface edge‑cases faster than any unit test suite.
Beta Testing Open Now
OpsFixer enters closed beta this month.
I built this app because it is something I wish I had throughout my career. Something that took a lot of the leg work out of reviewing, assessing and improving process maps.
Is it perfect? No. Will there always be some additional functionality to add? Yes. But, as a first step toward making process management, enhancements, and compliance reviews easier, it definitely ticks the right boxes.
OpsFixer Landing Page
If transforming static manuals into AI‑driven performance sounds useful, reply “OPSFIXER” to jump the queue. And if you enjoyed this build‑in‑public breakdown, subscribe to the AI for Leadership newsletter for weekly field notes on human‑centred AI.
If you want to try Lovable, some of the links on this page are affiliate links. They don't cost you any more but help support this newsletter.
Get anytime access to our growing collection of content; from usable prompts to help your leadership, useful articles to enhance your learning of AI and guides to using the technology that is going to change everything about how we work and lead our teams. ALL FOR FREE!!!