Building Two Production Platforms in 4 Months: What AI-Assisted Development Actually Looks Like at Scale

In late 2025, I joined a European sports organisation as their technology lead. They had a brand, a business plan, and a growing team. What they didn't have was any digital infrastructure. No website. No internal tools. No customer database. No analytics.

Four months later, they had two production web platforms, a customer data platform, role-based dashboards, investor reporting systems, push notifications, live scoring APIs, and more than 10 third-party integrations. All built by one developer using AI-assisted development.

This isn't a story about AI replacing developers. It's a story about what happens when you combine domain expertise, modern architecture, and AI tooling into a workflow that eliminates the friction traditional development teams spend most of their time on.

The scope

Let me be specific about what was delivered, because vague claims about "AI productivity" are everywhere. Here's the actual output:

Platform 1 - Corporate website. A CMS-powered marketing site with visual editing, dynamic team profiles, press coverage, venue information, and sponsor integration. SEO optimised, responsive, deployed to production.

Platform 2 - League Portal. A role-based internal platform serving five distinct user roles: players, captains, staff, executives, and administrators. 49 individual pages covering schedules, team management, feedback systems, surveys, announcements, profile management, and administrative tools.

Customer data platform. Importing and segmenting applicant data, subscriber lists, and user profiles. Export functionality. Activity tracking. Integration with the email system for targeted communications.

Referee platform. PIN-authenticated system for match officials to manage live scoring, game clocks, and match events during competitions.

QR check-in system. Team-specific QR codes for tracking attendance at events, with admin dashboards for monitoring participation.

Push notification system. PWA-based push notifications for announcements, schedule changes, and system alerts. Full subscription management.

Live scoring API. Real-time match data endpoints for integration with streaming overlays, scoreboards, and public-facing displays.

Content submission system. Player-facing content submission with admin review queues, moderation workflows, and publishing controls.

Investor reporting. Branded document generation for stakeholder communications, technology overviews, and product documentation.

Integrations. Sanity CMS, Supabase (database + auth), Stripe (payments), Microsoft 365, SharePoint, email automation, Google Analytics 4, push notification services, and several internal APIs.

All of this was built, tested, deployed to production, and serving real users within four months.

The architecture decisions that made it possible

Speed at this scale isn't about typing faster. It's about making architecture decisions that eliminate entire categories of problems.

Next.js with App Router. Server Components for performance, client components where interactivity demanded it. The App Router's file-based routing meant adding new pages was trivial - no routing configuration to maintain, no boilerplate to write.

Sanity CMS for content. Headless CMS with a visual editing studio. The non-technical team could update content, publish pages, and manage assets without touching code or waiting for a developer. This single decision removed what would have been hundreds of hours of content update requests.

Supabase for data and auth. PostgreSQL database with Row Level Security, built-in authentication (magic link, no passwords to manage), and real-time subscriptions. The RLS policies meant I could write data access rules once and trust them across every API route. No middleware authentication checks scattered through the codebase.

Vercel for deployment. Git push to deploy. Preview URLs for every branch. Edge functions. The deployment pipeline was effectively zero-configuration, which meant I spent zero time on DevOps and all of it on features.

Tailwind CSS v4. Utility-first styling with design tokens. Consistent visual language across 90+ pages without maintaining separate stylesheets or a CSS architecture.

Each of these choices was deliberately selected to minimise operational overhead and maximise development velocity. The entire stack is designed around a single principle: one developer should be able to build, deploy, and maintain everything without specialist infrastructure knowledge.

Where AI accelerated the build

AI-assisted development was not the architecture. It was the accelerator applied to the architecture. Here's where it made the biggest difference:

Component scaffolding. When the portal needed 49 pages, each with a consistent layout pattern but unique data and functionality, AI generated the initial component structure for each page. I'd describe the data shape and user interaction, and AI would produce the skeleton. My job was to refine the logic, connect the data, and handle edge cases.

Bug investigation across large codebases. When something broke, I could load the entire project into a single AI session and describe the symptom. The AI would trace through the relevant files - often six or seven components in the chain - and identify the root cause. One session found that a database column constraint was silently blocking an operation five files away from where the symptom appeared. Finding that manually would have taken hours of console logging and database inspection.

Database schema design. Describing the business requirements in natural language and having AI propose the PostgreSQL schema, including RLS policies, was dramatically faster than designing schemas from scratch. I'd review, refine, and test - but the first draft was usually 80-90% correct.

API route implementation. Supabase + Next.js API routes follow predictable patterns. AI generated these rapidly and accurately. Authentication checks, data validation, error handling, response formatting - the patterns are well-understood and AI handles them consistently.

Documentation and reporting. Investor documents, technical overviews, product documentation, internal guides - all drafted by AI and refined by me. The writing was good enough that refinement typically took 15-20 minutes per document rather than hours of drafting from scratch.

Where AI was not the answer

Security decisions. Row Level Security policies, webhook signature verification, authentication flows, session management - I wrote and tested all of these carefully. AI can suggest patterns, but security is an area where a subtle mistake can have serious consequences. Every RLS policy was manually reviewed and tested with real database queries.

User experience design. The portal serves five different user roles, each with different needs and different navigation structures. Designing that information architecture required understanding the organisation, its workflows, and its culture. AI can generate components but it can't decide what a captain needs to see versus what an executive needs to see.

Business logic. How should the points system work? When should notifications fire? What data should staff see versus what players see? These decisions required conversations with stakeholders, understanding of the organisation's goals, and judgment calls that no AI model can make.

Production debugging under pressure. When something breaks in production and users are affected, the debugging process requires a combination of technical knowledge, calm judgment, and domain context that AI assists with but can't drive alone.

The economics

Let's be direct about what this means financially.

A traditional development approach to this scope would typically require:

• 2-3 frontend developers
• 1 backend developer
• 1 DevOps/infrastructure engineer
• 1 project manager
• 6-9 months of development time

At market rates, that's roughly $300,000-500,000 AUD in development costs for a team, or 18-27 person-months of effort.

What was actually spent: one developer, four months, using AI-assisted development and a modern serverless stack. The infrastructure costs (Vercel, Supabase, Sanity) run to a few hundred dollars per month, not thousands.

This isn't a theoretical comparison. It's what happened. The platforms are live, serving real users, processing real data, and running in production.

What this means for businesses

If you're a business leader reading this and thinking "we need custom software but can't afford a development team," the landscape has changed.

The combination of modern serverless platforms (which eliminate infrastructure complexity), headless CMS tools (which give non-technical teams content independence), and AI-assisted development (which multiplies developer output by 3-5x) means that ambitious technology projects are now accessible to organisations that couldn't previously afford them.

The catch: you still need someone who understands architecture, makes good decisions, and knows when to trust AI and when to override it. The tools are powerful, but they're not autopilot.

If you're considering a platform build and want to talk through the approach, get in touch. This is the work I do at Cognitiv - and the case study above is exactly the kind of result I help other organisations achieve.