From Fields to AI: Why the Future of Intelligent Systems Isn't About Prompting Anymore

If in-house AI and vibe coding won't save your business, then what will?

When Johnny Morris was a teenager, he spent his days on building sites watching his father transform derelict monasteries and empty fields into something useful. The work fascinated him, but something fundamental became clear: the sheer capital intensity and risk made it difficult for individuals to succeed independently. So he ran towards technology instead, convinced he could accomplish more with data tools than with bricks and mortar.

Three decades later, Morris has spent his career proving that instinct right. He's built data platforms with Fortune 100 clients, created pricing engines for institutional real estate investors, and spent countless hours wrestling with the same problem: professionals drowning in information trapped inside PDFs and disparate systems. Yet his latest conviction challenges everything the AI industry is currently obsessed with. The way forward, he argues, isn't about learning to write better prompts or building your own AI system in-house. It's about understanding something far more fundamental: the shift from prompting to what industry insiders are now calling "context engineering". And if you work in professional services, real estate, or any knowledge-intensive field, you should care deeply about what that means.

The Unlikely Path to Fifth Dimension

Morris's journey to co-founding Fifth Dimension, an AI platform for real estate professionals, reads like a masterclass in strategic career pivoting. He began as a data analyst building automated valuation models at Hentra, then moved into product roles at major corporates like Countrywide. He launched a geospatial analytics platform with Fortune 100 clients and built out pricing engines for Way Home, a platform connecting institutional investors to single-family rentals. Each role crystallised a conviction: intelligent systems could solve more real-world problems than traditional infrastructure ever could.

But what he discovered across these roles was a persistent problem nobody seemed to want to solve properly. Real estate professionals lived drowning in data that wasn't actually data. Everything ended up requiring someone to read through a PDF. After years of wrestling with this friction point, Morris and Kate started Fifth Dimension three years ago, right as ChatGPT was beginning to reshape how people thought about AI.

They built with a specific philosophy from the start: intelligent technology should supercharge professional capability, not replace human judgement. The vision sounds straightforward. Executing it, he's learnt, is anything but.

The Messy Reality Behind the AI Hype

Walk into any organisation right now, and you'll hear some version of this conversation: We're spending 300K on Copilot, we're forming an AI committee, and we're considering building our own system in-house. Morris encounters these conversations constantly, and his response has crystallised into something close to a warning. Most people setting out on in-house development should give themselves six to eight weeks to learn what they're actually attempting. The gap between a compelling demo and a product that actually works for fifty people, five hundred people, or a thousand people hasn't actually narrowed. If anything, organisations systematically underestimate the challenge.

The Enterprise Search Trap

Take enterprise search as perhaps the clearest example. Every real estate business Morris has spoken to who tried implementing one has failed. The theory sounds sensible: plug into your SharePoint, your Box, your data repositories, and answer questions across all your documents. The reality is messier.

The problems multiply when the rubber meets the road:

• Scale and complexity mismatches: What works beautifully in a proof of concept with clean, curated documents breaks down when you introduce the actual mess of real organisational data

• Permission hierarchies and security: Enterprise systems carry 20 years of Microsoft permissions layers that create unpredictable behaviour and confidentiality risks

• Domain-specific knowledge gaps: New rent escalation clauses, unusual lease structures, or industry-specific practices the model has never encountered simply confuse the system

• Document variability: Some PDFs are 500 pages, others have coffee stains, some are scanned images rather than text, and the system struggles to handle this variability consistently
  

You can extract data from supplier contracts with Gemini 2.5. You can set up a beautiful proof of concept that looks impressive to stakeholders. But the moment you introduce actual leasing team data and ask people to use it in their daily work, everything breaks down. The problems are numerous and interconnected, requiring expertise that most organisations simply don't have.

Where Vibe Coding Reaches Its Limits

The mythology of vibe coding collides with production reality here. Yes, it's extraordinary for building wireframes and communicating ideas quickly. Morris's product team uses it constantly to mock up interfaces and test concepts with stakeholders. That's genuinely valuable.

But there's a profound chasm between a proof of concept and a production system. Building something that works seven times out of ten in a demo is not the same as building something that works reliably for hundreds of users across different workflows and edge cases.

Production requires what Morris describes as "really skilled engineers, specifically AI engineers, and really skilled kinds of real estate people with technology heads."

Most organisations don't have that combination. Most won't build it quickly. Yet the siren call of in-house development remains seductive, especially when the promise is total control and bespoke solutions tailored to your specific needs.

The Strategic Question Nobody Asks First

If there's one theme that runs through Morris's advice to real estate teams, it's the importance of asking the right question before you commit to a path. Not "Should we build this in-house?" but rather "What specific business outcome are we trying to achieve, and can we measure it?"

His framework for AI adoption follows a simple progression:

1. Find an early win: Identify one measurable use case that delivers quantifiable business value (reduced costs, accelerated timelines, new revenue) and solve it well with the best available tool, whether that's a third-party product or an external partner

2. Build credibility: That early win gives you internal capital to attempt more experimental work and justifies further investment from sceptical stakeholders

3. Create external commitment: Make a board-level promise about a specific outcome by a specific date, which proves far more powerful for driving change than internal enthusiasm alone

4. Invest in people, not just tools: Spend deliberately on training, change management, and expertise to help teams adopt and adapt the tools

   
   

Morris has noticed the quality of conversations with senior leaders has fundamentally shifted. Two years ago, they were discussing committees and theoretical plans.

Now, they're saying, "We tried all of that. We've learnt some stuff. But now we've got a vision. We know a specific part of our business we can transform. What's the gap?"

The ambitions have become concrete. The organisations advancing furthest are those who understand that the blocker isn't usually the technology anymore; it's organisational readiness and clear strategy.

Context Engineering: The Shift Nobody's Ready For

But here's where Morris's thinking gets genuinely forward-looking, and where Fifth Dimension's product roadmap is pointing. The next phase of AI development won't be defined by people's ability to write clever prompts. It will be defined by their ability to provide the right context to the system about the problem they're solving.

The distinction sounds subtle in theory. In practice, it reshapes what becomes possible.

Why Context Matters More Than Prompts

When you're using a generic AI tool for a generic task, you're starting from zero context every time. The system doesn't know your business, doesn't know your investment priorities, doesn't understand how a specific deal fits into your portfolio, and doesn't have memory of how similar situations were handled historically. That's inefficient. It's also why people feel they're constantly scratching the surface, convinced they're only using a fraction of what's possible.

But what if the system could learn your context automatically in the background? What if every time you wrote a valuation report, every time you underwrote a deal, every time you filed quarterly asset reporting, the system would build a contextual foundation that made it smarter next time? And critically, what if you had complete visibility and control over what that context was, able to edit it, remove it, add to it, or approve it?

That's where the technology is heading. Not invisible intelligence you have to trust blindly, but visible, auditable, controllable context that remains entirely in your hands whilst becoming progressively more powerful over time.

The Three Layers of Trust

Fifth Dimension's philosophy of transparency becomes essential here. The product is built around three specific layers that let users understand exactly what the system is doing:

• Reasoning: The system surfaces the reasoning behind every decision it makes, making the inner workings visible to the user rather than hidden

• Citation: The system shows what documents, memories, and specific context informed that reasoning, creating accountability and traceability

• Traceability: Users can click through to the exact source material, including specific passages from specific documents, eliminating the black box

  
  

If you disagree with the output, you can target your correction precisely: the reasoning was wrong, the source was irrelevant, or the information was simply inaccurate. That correction becomes a rule that governs future work. Over time, the system learns your domain, your priorities, and your standards without you having to explain them repeatedly.

The Contrast with Invisible Memory Systems

Compare that to how memory works in ChatGPT or Claude. These systems increasingly absorb information about users silently, and it's not always transparent when or why they're invoking what they remember. You discover it only when something unexpected surfaces, like a reference to a project you abandoned months ago, a detail from work you're no longer doing, or a misremembered priority that no longer applies. That invisibility creates friction and mistrust.

Fifth Dimension inverts the model entirely. Every memory retrieval is deliberately highlighted. Users can click through to see exactly what context was used and how it informed the output. The philosophy extends across the entire system. Understanding what context went in means understanding what's happening.

Why This Matters Now

The 2026 Inflection Point

We're at a critical juncture. The initial AI boom brought panic, enthusiasm, and a lot of expensive experiments that delivered modest returns. Organisations spent 300K on Copilot rollouts without changing how people worked. They formed committees but didn't clarify what they actually wanted to accomplish. Some started building in-house systems that would require years of maintenance against rapidly shifting foundational models.

Morris believes 2026 will reveal the gap between those who are on an actual transformation path and those still experimenting at the margins. The organisations moving forward have learnt that you don't need to build everything yourself. You need clarity on what you're solving for, investment in change management, and tools that remain intelligible and controllable.

Context engineering is the framework that enables this transition. It's less about individual user skill (learning prompting techniques) and more about system design. It's about building infrastructure so that less technical users can benefit from progressively smarter systems without having to understand the mechanics underneath.

For project delivery professionals managing complex portfolios, evaluating properties, and coordinating across teams, this shift matters deeply. The tools you'll be using next year won't just be faster versions of the tools you're using today. They'll have understood your business, learnt your priorities, and built context that makes them genuinely more useful.

The catch, as always, is execution. And execution requires clear-eyed thinking about what's actually necessary to make transformation real, rather than magical thinking about what's theoretically possible.

Worth Your Time

Morris's conversation covers far more ground than the handful of themes explored here. He walks through the genuine risks of enterprise search implementations gone wrong, explains why smaller contractors often outpace larger ones on technology adoption, and even shares what he's personally automating in his day-to-day work. There's a particularly telling story about how Fifth Dimension's original incarnation (an email-based assistant called Ellie) evolved into the agent backbone of the entire platform, guided by the philosophy that you retire products when the technology matures, not constantly, but you reinvent them instead.

If you're wrestling with how to think about AI's actual potential for your organisation, frustrated by the gap between hype and reality, or curious about what "context engineering" actually means in practice, the full conversation is worth your time.

Listen to the full Project Flux Spotlight episode with Johnny Morris to hear him discuss the decade-long journey that shaped his thinking about intelligent systems, why vibe coding is brilliant for POCs but insufficient for production, and why the shift from prompting to context engineering represents the next genuine frontier in AI adoption.