Building Without Code — How AI-Assisted Development Is Changing Web Projects

Over the past three years, a quiet revolution has taken place in how software gets built. AI-assisted development — using tools like Claude, GitHub Copilot, and other large language model assistants — has made it possible for people with business domain knowledge but limited formal programming training to build sophisticated, functional software. At Whamos, we have been doing exactly this, and the results have been remarkable.

What Does “Building Without Code” Actually Mean?

The phrase can be misleading. Building without code does not mean no code is written. It means the human in the process is not the one writing the code line by line. Instead, you describe what you want — clearly, specifically, and with real business context — and an AI agent produces working code based on that description. A skilled human then reviews, tests, directs corrections, and manages the overall architecture.

The skill that matters is not knowing how to write a for-loop. It is knowing exactly what you are trying to build, understanding how the pieces fit together, and being able to communicate requirements with enough precision that an AI agent can execute them correctly. This is a very different skill set to traditional programming — and one that business owners and domain experts can develop relatively quickly.

What We Have Built This Way

At Whamos and our sister company Technophobia Ltd, we have used AI-assisted development to build a range of products that would previously have required a full software development team and months of work. These include a Holiday Home Booking plugin for WordPress (now commercially distributed via Freemius with free and premium tiers), an Outdoor Pursuits Booking system for an activity company in the Lake District, a full Learner Management System for an adult education provider, a custom motorhome booking system for a Cumbrian car park operator, and our TP PageForge programmatic page building tool that is now used across all our client websites.

Each of these was built using a structured methodology: Claude for architecture planning and content, GitHub Copilot (which we call Wilson) for bounded code execution, and a clear system of project files, milestone journals, and sandwich prompts that keep the AI agents on track across long, complex build sessions.

The Methodology: Sandwich Prompts and Fresh Agents

One of the most important lessons from three years of AI-assisted development is that AI agents work best with tight constraints and clear context. We use what we call sandwich prompts — a Team Brief that sets the overall project context, a specific Task that defines exactly what needs to be done in this session, and Restraints that tell the agent what it must not do. This structure prevents agents from going off-script, making assumptions, or producing code that technically works but does not fit the broader architecture.

We also apply a fresh agent rule: every new milestone starts with a new agent session, loaded with the current project file as external memory. AI agents degrade in quality over very long conversation contexts — starting fresh with a well-maintained project file is consistently more productive than continuing an increasingly confused 10,000-word conversation.

What AI-Assisted Development Cannot Do

It is important to be honest about limitations. AI agents make mistakes — sometimes significant ones. They can write code that passes a casual review but contains subtle bugs that only emerge under specific conditions. They can misunderstand requirements if the brief is ambiguous. They can produce code that works in isolation but breaks when integrated with other systems. Human oversight, testing, and architectural judgement remain essential.

AI-assisted development also does not eliminate the need for understanding. If you do not understand what you are building at a conceptual level — the data model, the user journey, the integration points — you cannot effectively direct an AI agent or catch its errors. The human still needs to be the architect, even if the AI is doing much of the bricklaying.

What This Means for Cumbrian Businesses

The practical implication for small and medium-sized businesses in Cumbria is that bespoke software is now within reach in a way it was not three or four years ago. Systems that would previously have cost £50,000–£100,000 to commission from a traditional software house can now be built for a fraction of that cost — particularly when the business owner can actively participate in the specification and direction of the build.

At Whamos, we have documented our methodology in a six-volume series called Building Without Code, which records how each of our major software products was built, what decisions were made, and what we learned along the way. It is both a record of our process and a practical guide for others who want to explore AI-assisted development.

The Future Direction

AI coding tools are improving at a rapid pace. The quality of code produced, the ability to maintain context across long sessions, and the sophistication of reasoning about architecture are all significantly better than they were even twelve months ago. We expect the gap between what a skilled AI-assisted developer can produce and what a traditional development team can produce to continue to narrow.

For businesses in Cumbria and beyond, this means more accessible bespoke software, faster development cycles, and the ability to iterate and improve systems without the overhead of large agency retainers. If you have a process that you think could be improved with a custom software solution, we would be glad to talk through what is possible.