People are still obsessed with prompt engineering, and whilst it’s absolutely critical, they’re missing the bigger picture. They spend hours crafting the perfect instructions, attend workshops on “advanced ChatGPT techniques,” and think they’re winning because they can get Claude to write better responses to an overflowing inbox.
A good prompt is only half the battle. Individual prompts matter enormously – they’re the foundation of everything. But they only reach their potential when they’re executed in environments where the right context is available. That’s what most people are missing.
Playing at the wrong level
We’re not in 2023 anymore. Everyone’s asking “How do I get better at AI?” when they should be asking “How do I create the conditions where my AI interactions actually work better for me?”
I keep seeing this with the teams we work with – they start excited about crafting perfect prompts. They’ll spend ages getting Claude to understand exactly what they want. That’s not wrong – it’s essential. But brilliant prompts are only the starting point.
Those that have really nailed AI have gone a level deeper. They’ve built systems and infrastructure where every prompt – no matter how simple – has access to the context it needs to be genuinely useful. Your prompt engineering skills still matter massively. But the same prompt that gives you decent results in a vacuum becomes transformative when it can draw on your actual knowledge, processes, and data.
Case in point is a financial services team we worked with a few months ago. Initially, they were crafting increasingly sophisticated prompts for Claude to analyse compliance reports. The prompts were genuinely excellent – detailed, specific, well-structured. But they were still starting from scratch every time – explaining context, providing background, feeding in external data to refer to.
Then they tried something different. They built an environment where Claude could access the data it needed. It had an overview of previous conversations and outcomes. Suddenly, even their basic prompts started producing results that were better than their sophisticated ones had been because the AI had the context to understand what mattered.
That’s when I realised what’s actually happening here. We’re all becoming digital gardeners.
Building environments, not just crafting prompts
Good managers don’t just give perfect instructions. They create conditions where even simple directions lead to excellent outcomes. And I see this exact same behavioural loop happening with AI.
It makes sense – you wouldn’t hand off a task to a human being without context. “Oh, could you please draft up a contract for me with a supplier?”. Who is the supplier? What is the agreement? What has worked well in the past?
The best prompt in the world struggles without context. Most AI users are stuck in a behavioural loop that feels like it works. But it’s time-intensive and they are still operating in isolation. They craft beautiful prompts, review every output, provide context every single time. The prompts work well, but they don’t compound.
Prompt engineering gets you in the game. Environment design is what wins it.
What makes a thriving digital garden
- Just like actual gardens, AI systems need the right environment for individual interactions to flourish:
- Knowledge bases that make sense – Information AI can actually navigate and use, so your prompts don’t need to start from zero every time
- Clear examples of good work – So even a simple “do this task” prompt knows what you’re aiming for
- Your actual processes written down – Not just what you think you do, but what you actually do, so prompts can build on established workflows
- Context about your organisation – The unwritten rules and cultural stuff that make the difference between generic and genuinely useful responses
This is what good managers have always done for human teams. Create conditions where even basic instructions lead to great outcomes.
Getting started
The sophisticated digital gardeners I know have AI that can access company knowledge, follow documented guidelines, connect with calendars and email, understand team dynamics, and know when simple prompts need complex context.
This can be broken down into four clear steps:
- Map what you know – Where does information live? How structured is it?
- Write down your decisions – What rules do you follow? What makes work good?
- Spot the patterns – Which workflows repeat with minor variations?
- Connect the pieces – How can your knowledge talk to your AI tools?
The problem for leaders
Here’s what’s going to happen. Some teams in your organisation will figure out how to build these digital gardens. Others will keep perfecting their prompts in isolation.
Both groups will have excellent prompt engineering skills. But the teams building digital gardens will get transformational results from the same quality prompts while everyone else is still crafting perfect instructions that lack context. This will become repetitive and time intensive – two of the pain points AI promises to eliminate.
Don’t stop getting better at prompts. But start building environments where your AI interactions can reach their actual potential.
