I recently sat down with Brian Madison, the brains behind BMad — a breakthrough open-source framework for agile, AI-driven development, which has become a go-to resource for developers who want structured, repeatable workflows without handing over their thinking to an AI.

Brian’s approach very much clicks with how we think about AI at Nearform. We don’t view AI as a layer, it’s the way we build — and that’s what helps enterprises move from AI pilots and slow delivery to production-grade intelligent systems.

As we spoke, we quickly realised we agreed on one fundamental thing: AI agents don’t understand ‘vibes’. They need precision, crystal clear specs, explicit instructions and defined tech stacks. When you give them those elements, transformation unfolds. As a developer, you get less rework, faster shipping, and code that actually does what you need it to do.

That’s what the BMad method helps developers to achieve, and it dovetails with our approach at Nearform.

Below are some highlights from our full conversation. But we did also record a quick-fire Q&A at the end, which you can watch here:

https://www.youtube.com/watch?v=6158efbowgg&feature=youtu.be

I’ve talked a lot about spec-driven development (SDD). BMad is one of several tools driving SDD inside Nearform. Clear specifications force you to make decisions up front, guiding every step, uncovering assumptions and ensuring alignment with the original intent.

Many devs know that BMad is the actionable methodology that brings SDD to life. It provides structured guidance, defined workflows and even personas, to shift the software development process from “we have a vision” to “here’s exactly what we’re building”. It’s one of the engines of choice driving SDD, making it practical and incredibly effective for complex, AI-native development.

How did we go from Brian the person to BMad the approach?

Brian’s BMad journey

Brian is a software engineer, with over 20 years’ experience, who was, like many of us, trying to get AI agents to build complex apps autonomously. He hit a wall, where the AI started doing “stupid things”. Not in a malicious way, but in an ‘I misunderstood what you actually wanted’ way.

During one particular late-night experiment to fix the problem, it clicked for him. He realised the agents were missing context and clear guidance. And once he fed those elements into the system — through PRDs, defined tech stacks and explicit instructions — the quality of output improved massively. He’d built a highly-structured system, which enabled a deep, shared understanding between the expert and the AI. In doing so, it ensured everything produced in lines of code laddered clearly back to the defined spec.

That solution — the BMad method — has now grown into one of the most widely-used open source AI development frameworks. The open source nature of BMad also very much reflects Nearform’s approach — open source is in our ‘DNA’, it’s fundamental to how we work.

Brian built BMad to treat agents like facilitators, not oracles. Human experts bring the knowledge and the context, an agent asks questions to get clarity and then builds a solution that meets the defined business need. That’s the approach we embrace at Nearform.

Ultimately, it ensures every line of code is traceable to a single, machine-readable specification. This eliminates “AI drift”, speeding up the development process and minimising rework and technical debt.

What’s next for BMad?

First things first, we got confirmation that the much-loved BMad personas — Mary, John and Sally — aren’t going anywhere, especially after Brian confessed about the community backlash he received when he once tried to change the names.

We also got into what Brian has in mind next for BMad (and it’s not ‘Bhappy’). He made clear there’s a lot of momentum behind V6 of the methodology, the skills architecture and dev loop automation.

There’s also been a lot of talk around the development of a UI or dashboard for BMad, essentially a front end for the people who don’t want to touch a terminal, or less technical users like project managers — and it’s something Brian’s seriously considering.

Increasing predictability means amplifying AI success

Here’s the thing Brian and I kept coming back to: when engineers have clear specs and structured guidance (when they stop guessing and start building), everything shifts. They can focus on the hard problems, not the back-and-forth. The AI handles the build, they define the strategic guard rails.

We’ve learnt a lot from Brian’s work, and we’re watching closely as other frameworks and approaches mature in this space. What matters to use at Nearform is the discipline: defined specs, explicit instructions and traceable outputs.

Curious how AI-native engineering and SDD can transform your enterprise and accelerate your success? Get in touch.

Individual tools are great, but they’re not enough. Real impact comes from industrialising AI.

Look, AI is absolutely transforming software engineering right now. But here’s the thing… handing everyone Copilot and calling it a strategy is like giving someone a Ferrari and expecting them to win Formula 1. Individual tools are great, but they’re not enough. Real impact comes from industrialising AI.

We recently hosted a webinar: “Copilot isn’t a strategy: how to industrialise AI in the enterprise”, with some seriously clever people. Dave Kerr (Head of Engineering at Quantum Black Labs, McKinsey), Katie Roberts (Technical Director here at Nearform), and I joined Helene Haughney (Global Accounts owner & SVP for Global Delivery Transformation, Nearform) to dig into this properly. For the full deep dive, you can watch the full video below. But I also wanted to pull out some of the key points that I reckon matter most.

https://watch.getcontrast.io/register/nearform-copilot-isn-t-a-strategy-how-to-industrialise-ai-in-the-enterprise

The elephant in the room: Individual gains don’t scale

Dave spoke about the revolution in software engineering currently happening. The real shift happening right now is that AI is transforming how we do software engineering itself, moving towards what Dave called “software engineering factories”. Of course, individual engineers are seeing productivity wins. Nearform’s poll during the webinar showed that 68% of attendees are seeing gains. That’s brilliant. But here’s where people get it wrong: those individual supercharges don’t automatically translate to organisational impact.

The honest truth? You can hand out Copilot licenses tomorrow and watch engineers code faster. But you still end up with the same bottlenecks, the same broken processes, the same burnout. You need to scale that individual productivity into a collective force — build AI-native engineering teams that can deliver actual systemic change. That’s the real challenge.

Welcome to the era of “big code” (and it’s complicated)

We discussed the key challenges that are emerging from this new era of “big code” and asked how it changes our approach to development. Dave absolutely called it out during the webinar, and, in my opinion, he he nailed it:

“The honeymoon is over. Individual team members have superpowers, but collaboration is actually getting harder and harder and harder.”

He then asked the question we were all thinking: “If the honeymoon is over, what is marriage counselling 101?”

We came up with the following answers:

• Create time and space to learn this new world. Stop treating agentic engineering like a side project. Engineers need to actually experiment, try different approaches and figure out what works for their team.
• Build your machine. Your dev environment is your most important product. Automate everything. If you’re doing something three times, it’s automated.
• Work on your relationships… with your team members and with your LLMs. Your job isn’t to review code anymore, it’s to build feedback loops. So, when something doesn’t work at first, you improve the system, the sub-agents, the skills, the processes and the collaboration.
• And communication? That’s the golden skill now. Individuals can be colossally fast and productive, but that doesn’t scale. We need to collaborate, communicate and find ways to make sure other people can actually work with what we’re building.

Dave’s describing what at Nearform we’ve been calling the era of “big code”. The successor to “big data”, it implies massive velocity, huge volume and insane variability. And here’s a concrete example: Dave mentioned a 20,000-line pull request. You know what a human trying to review that is? Broken. It’s impossible. And it’s demoralising.

What happens? Burnout. Cognitive overload. Engineers who were flying three months ago now feel lost. At Nearform, we’re clear on this: individual productivity gains don’t automatically equal whole-team productivity. You need a fundamentally different operating model to handle this scale and complexity. Otherwise, you’re just running faster towards a wall.

Brownfield is the reality check we all need

Many enterprises face the challenge of implementing AI in complex brownfield environments. How do we tackle this and ensure safeguards in regulated industries?

Brownfield is real. It’s messy. And it’s where most of us actually live, right? Katie’s advice during the webinar was spot on. She said, “Don’t boil the ocean”. Instead carve out smaller spaces, write clear specs, be very explicit about your intent with your LLMs and set rules around it.

Essentially, don’t try to AI-enable your entire legacy codebase tomorrow. That’s a recipe for disaster. Instead, isolate specific areas, strangle them out (in the technical sense) and pick spaces where you can build new features or tackle real technical debt with clear intent. Small, intentional moves. That’s how you move the needle without imploding.

Trust stack: Because velocity without safety becomes chaos

With AI generating so much code, how do we ensure robust safeguards and maintain accountability, especially in regulated industries?

Agents can’t be held accountable when things go wrong. That’s on us.

So safeguards aren’t optional, they’re non-negotiable. At Nearform, we’re building a ‘trust stack’ in parallel to the tech stack, integrating layers of rigour across the entire SDLC:

• Static analysis to catch obvious issues
• Agent-based code reviews that actually understand intent
• Eval suites to validate behaviour
• Formal methods where it matters
• AI red teams for threat modelling
• Clear accountability chains so everyone knows who’s responsible for what

The goal here is simple — velocity increases, but stability and security never take a hit. You want to move fast? Grand. But not at the cost of trust.

Industrialising AI in your enterprise isn’t optional anymore, it’s how you stay competitive. And while it’s not straightforward, it’s absolutely doable. It starts with being honest about where you actually are right now. Then, it’s about creating a clear roadmap and working with a partner who’s actually done this in the real world. Not outsiders talking theory, but the practitioners shipping it.

At Nearform, we’re on the front lines — navigating brownfield complexity, shipping robust safeguards and scaling teams to become agentic factories.

It’s the enterprise engineers who should be the heroes here. For those organisations ready to move past one-off productivity gains and unlock what AI can actually do at scale, get in touch.

In the meantime, our Impact Talks series continues with the next webinar, The AI execution gap is opening. Which side of it are you on? on June 16th. Register here to secure your spot to learn how to be on the winning side of the gap.

Businesses aren’t short on AI ideas — they’re drowning in them. Discover how to cut through the noise, identify which AI opportunities deliver real value, and build a clear path from experimentation to enterprise-wide scale.

AI hasn’t created an ideas shortage. If anything, it’s created the opposite.

Businesses are surrounded by a huge range of potential AI use cases, with opportunities to improve customer service, accelerate internal processes, support employees, enhance decision-making, reduce operational costs and create new digital products. Vendors are bringing new ideas, internal teams are experimenting, and execs are asking what AI can do for growth, productivity and competitive advantage.

But this has left leaders unclear on which AI opportunities will deliver meaningful value, how they can prove that value quickly, and how they can scale the solutions across the enterprise.

From AI activity to AI maturity

The overwhelming potential for AI opportunity has actually seen many AI efforts start to slow down — not because the ideas themselves are weak, but because the ambition behind each one is, and the tieback to meaningful business value isn’t clear.

The real challenge is often more structured. Enterprises lack a repeatable operating model for deciding where AI will create value, proving that value at speed, and then scaling what works. They may already have experiments running, proof-of-concepts in progress and teams exploring new tools, but — rather importantly — AI activity is not the same as AI maturity.

AI maturity isn’t about how many tools you’ve adopted, or how many pilots you’ve launched. It’s about whether you’ve moved consistently from opportunity to outcome. In simple terms, it sees organisations follow a repeatable journey: idea, proof, scale and repeat.

Each stage requires a different level of discipline. At the idea stage, the challenge is prioritisation. Enterprises often have too many possible use cases, and not enough clarity on which are most closely tied to revenue, cost, risk, customer experience or strategic advantage.

At the ‘proof’ stage, the focus shifts to validation. A proof-of-concept may show something that’s technically possible, but that doesn’t mean it’s valuable, usable or worth scaling.

At the ‘scale’ stage, it’s all about production. AI solutions need to fit into real workflows, integrate with existing systems, meet operational requirements and be adopted by the people and teams they’re designed to support.

Finally, at the ‘repeat’ stage, the challenge is capability. Enterprises need to avoid starting from scratch every time, so the focuses here are patterns, governance, reusable components, delivery discipline and a clearer way to manage a portfolio of AI opportunities.

This is the maturity gap, where we move beyond curiosity and turn AI into a consistent source of business value.

Execution is becoming a differentiator

To try to prove value, enterprises rely on AI pilots as they help explore possibilities, test assumptions and build confidence. But while they are important, they’re also not the finish line.

Too often, pilots aren’t designed with production in mind. They sit outside business processes, measure success in technical terms rather than commercial impact, or lack a shared execution model. This results in promising ideas becoming fragmented efforts with unclear ROI, while the pressure to show faster process remains. And leaders feel the tension — they know AI matters to remain competitive, but they also need to avoid wasted investment and disconnected experimentation.

The answer isn’t to experiment less, but to focus on more purposeful execution. It’s identifying the right opportunities and turning them into measurable business outcomes with speed and discipline. This requires a different way of working. Enterprises need a way to move quickly without things becoming chaotic, and ideas need to be tested without losing sight of production.

Through AI-native engineering, teams can explore, prototype, build and iterate faster than before, creating more room for experimentation. But this approach raises the importance of focus. If organisations can run more experiments, they need a stronger operating model to decide which experiments matter, what outcomes they should target and which solutions they should scale.

This is where an AI factory comes in.

Moving from experimentation to a robust AI operating model

To move from activity to impact at pace, organisations need a repeatable operating model. An AI factory provides this structure, connecting business priorities with delivery, adoption and measurable value. Instead of one-off projects, it creates a consistent path from idea to outcome.

Now, the word ‘factory’ can sound mechanical, so it’s worth being clear about what we mean in this context. It’s not about mass-producing generic AI solutions, but about creating a repeatable way of working that helps enterprises connect AI ambition to business priorities more quickly.

A good AI factory brings together the ingredients that are often fragmented across an organisation: business prioritisation, product thinking, data capability, AI-native engineering delivery, governance, adoption planning and value measurement. It creates a structured path for deciding what opportunity to pursue, how to test them and how to scale them if the value is real.

In a nutshell, an AI factory ensures every initiative starts with the right problem, proves value quickly and is built to scale.

At Nearform, we use the 3–3–3 rhythm to illustrate this: 3 days to prioritise, 3 weeks to prove value, 3 months to launch a first release.

This doesn’t suggest every AI opportunity is identical, or that every organisation follows the same path. The shape of the work using this approach will always depend on the business problem, data environment, technical complexity, operating context and appetite for change. The point of the AI factory rhythm is simply to create momentum, while keeping the work anchored in value.

1. The first stage, prioritisation, means asking questions like: which opportunity should be pursued first? Which is tied to a meaningful business outcome? Which has the right combination of value, feasibility, data readiness and organisational sponsorship?
2. The second stage, proof , looks at whether the opportunity can be validated quickly and whether the team can test if the solution will create value for users and the business.
3. In the third stage, the focus is launch . Can the solution be taken into a real environment, integrated into workflows and measured against the outcome it was designed to improve?

Building confidence through structure

For many leaders, one of the hardest parts about AI transformation is telling a clear and confident story about progress. Leadership teams want to know where investment is going, business units want to understand how AI will help them to achieve their goals, technology leaders want to ensure solutions are robust, scalable and responsible, and teams want clarity on what to work on next.

A factory model helps by creating structure. It gives leaders a way to manage AI as a portfolio of opportunities, rather than a collection of disconnected experiments. It helps teams navigate a common path, from idea to proof to scale. And it makes it easier to see which initiatives are working and which need to change.

As AI-native engineering increases the speed and capacity of delivery, the need for a clear AI operating model becomes even more important. The organisations that win won’t necessarily be those with the largest number of AI pilots, they’ll be the ones that build the maturity to turn the right ideas into value, and then scale that value with confidence.

At Nearform, we work closely with enterprises to get their very own AI factory up and running. Ready to find out more? Contact us.

We’re answering some of the most common questions around SDD.

Building with AI is an iterative process — discovery, testing, and refinement are part of the work. But most teams are building critical software on a mix of tribal knowledge, ticket fragments and best guesses.

Spec-driven development (SDD) fixes that. It forces clarity upfront, creates reviewable specs in plain markdown and turns them into clean tasks and reliable code. AI does all the heavy lifting, but humans stay in control. This creates a robust workflow, minimising ambiguity and rework, while allowing teams to ship complex systems with far more confidence.

Below, we’ve answered some of the most common questions around SDD.

What is spec-driven development?

SDD is a software methodology that places emphasis on writing the spec and intent in a guiding document before the code. Once intent is defined, it leads both humans and AI through the whole software development process, from planning, through to task breakdown, through to implementation and scale.

SDD is a methodology, rather than a single tool. It produces reviewable markdown specs, iterates on them with a human in the loop, then, depending on the tool, it turns approved specs into an implementation plan and code, reducing ambiguity, highlighting edge cases and producing a process that can scale easily across teams.

Ultimately, you capture what you want to build, and how it should behave, then the model builds against that agreement. AI helps with planning, decomposition and drafts, but humans remain accountable at every step.

Why is SDD gaining traction?

AI needs clarity. Coding agents are powerful, but they can’t fully understand “vibes”. They need precise, unambiguous instructions, and that’s exactly what SDD helps software teams to deliver.

The bottleneck for many teams isn’t typing code, agents can spin up hundreds of lines of code in a matter of minutes. The real bottleneck is not having a shared understanding of detailed requirements, constraints, necessary trade offs and edge cases. That’s where SDD focuses effort.

Lastly, many organisations are struggling with scaling AI across the enterprise. SDD directly addresses this challenge. Specs become the single source of truth, allowing teams to onboard faster, split workstreams cleanly and avoid duplicated work or conflicting assumptions. It also makes auditing and other aspects of risk mitigation more efficient.

When should I use SDD?

While it’s an incredibly powerful methodology, SDD isn’t always right for each and every scenario (or developer!). Some of the scenarios it fits best include:

• Delivering greenfield projects with a clear objective and stakeholders that are available to shape a first-class spec
• Creating mid-sized, well-bounded features in brownfield systems where integration points are known (for example, a new workflow, an API surface or a reporting module). In some brownfield projects, it may initially require slightly more work, but the benefits are still visible early in the process.
• Delivering in regulated or high-risk domains, like healthcare or financial services. Human in the loop review at every stage, alongside robust audit trails, means SDD keeps quality and traceability high.

We’re aware that many feel SDD is ‘less experimental’ compared to vibe coding, but you can also run experiments and technical spikes to test assumptions to quickly test ideas. Then your learnings can be crystallised and summarised by writing a proper spec, allowing it to be a starting point for the actual production-ready implementation. This is what we’re doing in multiple teams and it’s working efficiently.

Additionally, for projects with massive legacy codebases without clear seams, SDD might require preparatory work to build project documentation to create boundaries.

What outcomes should I expect from SDD?

When relying more on specs, the intent and the “why” of a specific change gets captured. This is important as when the system grows, it becomes more vital to know why a feature is behaving in a specific way. As such, there are also a handful of extra benefits SDD brings for CTOs, including:

• Fewer surprises. Defining explicit specs at the outset exposes any assumptions early (whether it’s technology choices, non-functionals or edge cases) and provides transparency for any future audits.
• Better forecasts. With SDD, work is decomposed into reviewable tasks, this makes tracking progress and risk much easier.
• Higher consistency. Teams coordinate around workstreams, rather than individual tickets, which reduces collisions and rework.
• A cultural shift. SDD allows developers to spend more time on context engineering, articulating the “why” and the “what”, and less time on line-by-line craftsmanship. This delivers clear payoffs in throughput and reliability.

What’s the business impact of SDD?

Let’s give you an example of a client engagement where Nearform recently applied the SDD methodology. For a leadership advisory firm, we took a strategic, legacy reporting platform and — using AI-native engineering with spec-driven development — we shipped a six-month scope in just over eight weeks. This was four times faster than a competitor’s estimate, and it cut report generation from days to hours.

How does SDD work?

SDD’s lifecycle activities are as follows:

• Define the brief — everything starts by capturing intent, constraints and success criteria, right from the outset.
• Initiate AI-assisted planning — then you generate a draft spec in markdown and iterate it until it reflects your true goal.
• Decompose specs into tasks — once it’s approved, the spec is then turned into an implementation plan.
• Implement in slices — you can then build the system, task-by-task, keeping tests close to the work.
• Review and ship — AI removes the toil, humans keep the accountability, reviewing every spec and code change.

SDD isn’t about simply pressing a button and getting a new app. It’s a human-directed workflow that uses AI to help teams move faster, with more confidence.

What skills should I expect from my developers for SDD to work?

For SDD to work effectively, particular skillsets are most effective. A senior developer needs to not only have rigorous technical ability and specialised architectural understanding, but they also need to have strong communication and critical thinking skills, as well as the ability to be ‘product-minded’ given focus shifts from writing code line-by-line, to orchestrating agents and defining intent.

C-suite leaders might not be used to developers having a strategic voice. But it’s important to recognise that with SDD, engineers become the architects of business logic. Hierarchical organisations that treat developers as merely assembly line workers will ultimately struggle to leverage AI effectively.

Running on vibes is fun… until it isn’t. SDD brings the clarity that teams secretly crave, and the pace they didn’t know they could hit.

If you want a taste of what working from a tight spec actually feels like, we’re right here.

Slow, committee-based governance is being replaced by AI-native systems that enable faster, more transparent, and better-informed decision-making.

For decades, highly-regulated sectors have relied on layers of approval and committees to manage risk. This model was originally built for safety, but over time, it’s calcified into operational paralysis. And in today’s market — where AI‑driven automation, real‑time decisioning and constant iteration set the competitive pace — that old model is slow and no longer fit for purpose. This is evident across many sectors, from banking and financial services, to healthcare and insurance, to telecommunications, and more.

The era of ‘decision by committee’ is ending and the age of AI-native governance is beginning.

When safety becomes stagnation

In highly regulated environments, risk aversion often blocks agility.

Traditional operating models were designed for a world of predictable processes and linear outsourcing. Organisations were designed in ‘pyramids’, with junior analysts cleaning data, middle managers reviewing reports and senior leaders making decisions on data that’s already days old. However, this junior-weighted structure isn’t able to keep pace with the speed of modern demands. In fact, what it creates is a system where safety measures actually increase operational risk by delaying — or even preventing — timely reactions to change.

A recent survey from McKinsey evidences this strong relationship between seniority and maturity when it comes to governance and compliance, as organisations with lower-ranked heads of compliance consistently score themselves lower on maturity.

The shift to AI-native decision making

AI-native engineering changes the catalyst of productivity from ‘human talent’ to ‘expert + AI’ outputs.

Nearform’s AI-native engineering approach is grounded in 15 years of software development lifecycle (SDLC) expertise, across over 700 client engagements. What this experience has taught us is that the answer isn’t simply adopting more AI tools, but instead delivering scalable frameworks and governed ways of working that allow AI agents to operate systematically alongside humans.

Ultimately, agentic technology allows for the collapse of the traditional ‘pyramid’ model. Instead of a 10-person team managed by layers of review, a single, senior expert orchestrates a structured team of specialised AI agents that operate within automated safety frameworks and guardrails.

In AI-native engineering, agents do more than simply automate tasks. They also participate in the core workflows that determine speed, alignment and risk across the entire software lifecycle.

Agents ingest data, update models and run scenarios in real time, surfacing risks and drafting specifications, without any lag from manual handoffs. While doing so, every action is logged and replayable.

Ultimately, the human expert frames questions, sets constraints and applies judgement, while AI handles the execution. This enables faster decision cycles and broader coverage, with vastly consistent quality. In fact, in one recent client engagement, we were able to reduce the pivotal discovery phase from six weeks to just two weeks.

Fear vs. reality

There’s a natural hesitation around letting AI into key workflows, as there’s a persistent fear that agentic AI is unpredictable and uncontrollable. But the reality is not so bleak.

In practice, human-run processes can be inherently opaque and prone to inconsistent execution. It’s natural that people forget steps, improvise under pressure and hide errors. As an example, you can’t ‘replay’ a committee meeting to see exactly how a decision was reached. AI-native systems, by contrast, are more like ‘glass boxes’ — they leave audit trails, can operate within strict, human-defined programmatic guardrails infrastructure and, in so doing, strengthen compliance, as policy enforcement is embedded directly into the code, rather than being bolted on as an afterthought.

As a parallel, a human chef might create a new dish based on instinct and experience — you can’t then replay the exact steps to replicate the meal in exactly the same way. However, an agent chef would work from a recipe, where every ingredient, measurement and step is documented and reproducible.

What’s also important to note is that agentic systems can be restricted by intent validation gates and don’t have independent access to power. Before they can do anything, agents need APIs, credentials, compute power, network access and permission to use tools, all of which sit in human-built infrastructure. While an AI-native system cannot manifest physical power, it can only acquire digital resources from renting cloud computing to transacting via digital wallets if its objective is too broad and its guardrails are not programmatically defined and enforced.

The shift to AI‑native governance

The real change happening inside regulated organisations isn’t about engineering, but it’s about how decisions are made. AI‑native governance replaces slow, people‑only decision chains with systems where humans and AI operate together inside transparent, audited, policy‑driven frameworks. Instead of waiting for committees to meet, review, and sign off, organisations can move to real‑time, evidence‑based decision cycles where risk is surfaced instantly and action can follow without delay.

Now, this doesn’t mean AI takes decisions away from people, it just helps to advance the quality of the decisions people make. Intelligent systems can analyse vast datasets, highlight inconsistencies, flag emerging risk, and simulate outcomes in seconds. Humans then apply judgement, context, and accountability on top of that insight. This shift transforms governance from a reactive function to a proactive one.

This also means that under AI‑native governance, the role of our leaders will change. Instead of reviewing stale reports or second‑guessing incomplete information, leaders get a continuously updated view of risk, performance, and compliance. They move from gatekeepers to orchestrators — setting intent, defining constraints, and shaping guardrails that AI systems execute against. It also means that every action taken by those systems is logged, traceable, and replayable. This level of transparency simply doesn’t exist in human‑only governance.

What does this mean for decision‑making? Decision cycles that once took weeks shrink to minutes. Risks that were previously invisible, are surfaced instantly, and the quality, consistency, and defensibility of decisions increase dramatically.

This is what replaces the ‘decision committee’ — not unchecked automation, but trust‑centred, AI‑native governance that strengthens oversight while removing delay.

Are you ready to say goodbye to the ‘decision committee’ and trade paralysis for precision? Get in touch - we’re ready to help build your trusted AI-native future: https://nearform.com/

AI‑native engineering is a fundamentally different way of creating software — one where automated code generation, AI-driven agents and continuous learning systems reshape everything from architecture to workflows to economics.

We’re standing at another industry‑defining moment. When Agile first emerged, it injected raw urgency and optimism into software delivery — a belief that teams could build better, faster and with far more ambition. It reshaped the landscape, and the industry never looked back. But Agile is no longer the source of differentiation it once was. With nine in 10 businesses now practicing it, almost everyone is moving at the same pace, making advantage marginal, rather than material.

Today, a shift even more profound is underway. But this isn’t a new methodology or a rebranded process. AI‑native engineering is a fundamentally different way of creating software — one where automated code generation, AI-driven agents and continuous learning systems reshape everything from architecture to workflows to economics. It dismantles long‑held constraints. It accelerates value creation. And it rewards organisations that move decisively, meaning that those that don’t are falling behind, watching a new competitive order form without them.

This is why AI-native engineering demands a fundamentally different mindset. Classical software engineering has 40 years of deterministic compute and ironclad estimation models. If you’ve built something similar before, you can predict outcomes with reasonable accuracy. AI-native systems are different: they’re probabilistic and non-linear. Run your model today and it’s perfect. Run it tomorrow and it hallucinates. Progress isn’t a straight line… it’s a zigzag. That means estimation models that worked for decades no longer apply. You can’t tell your CFO what you’ll build in a year or how long it will take. You’re asking finance to embrace uncertainty on a technology they don’t understand. You’re asking for a leap of faith.

This is why early pilots often fail, not because the technology doesn’t work, but because we measure outcomes using metrics built for deterministic systems.

The uncompromising catalyst

Agile changed everything. When it emerged, it broke enterprises out of heavy, linear delivery models that couldn’t keep pace with digital demand. It brought faster cycles, tighter feedback loops, empowered teams and a new belief that software could (and should) evolve continuously. For years, that shift created enormous competitive advantage, and the organisations that embraced Agile early became the ones who shipped faster, learned faster and adapted faster.

Now, nearly every enterprise operates with some form of Agile delivery, yet many still struggle with slow releases, sprawling backlogs, dependency bottlenecks and legacy systems that simply can’t move at the pace the market demands. Agile optimised the process of human-led software development, but it didn’t change the unit economics or the underlying constraints. Agile is now simply best practice. Everyone does it. Competitive advantage disappears when your competitors are using the exact same playbook. AI‑native engineering becomes the new frontier, where differential advantage is created — just as Agile did in the mid‑2000s.

This is where the parallel begins. Just as Agile once redefined how teams worked, AI‑native engineering redefines what teams can make. It replaces manual, people-only delivery with AI‑accelerated creation, automated orchestration and systems that evolve continuously. Agile sped us up, but AI-native engineering changes the physics entirely.

It’s the uncompromising catalyst for colossal change, forging smaller, faster and truly nimble companies. AI-native engineering works on continuous learning loops, so software no longer ‘releases’ and stabilises, but it evolves. When systems adapt based on usage, performance and feedback, capability compounds over time. It also allows for data to be deeply integrated into the development process, meaning architecture, delivery and decisions are driven by real‑time organisational data, enabling outcomes to improve continuously as the system learns. With specialised agents handling complex tasks — from documentation and testing to infrastructure configuration and optimisation — your people can focus on high‑value problem‑solving. Such AI also generates, tests and refactors code continuously, allowing systems to improve at a pace traditional teams can’t match.

All of this points towards unmatched productivity and power. In fact, early enterprise adopters report a 20% productivity lift across development and service functions.

Agile’s legacy. AI native’s reign.

AI-native engineering is a total shift that’s bulldozing traditional moats. Sprawling legacy codebases, for instance, are suddenly not the burden they once were. AI can refactor, optimise or even outright rewrite them. Institutional knowledge, usually locked away in silos, is now easily codified and leveraged by agents. Even prized data gets supercharged by AI, driving insights and actions at a scale we’ve never seen before.

AI-native startups are adopting these high-velocity models from day one, leaving everyone in their wake. That’s creating a brutal, structural disadvantage for anyone still clinging to the old ways. Take a look at McKinsey’s State of AI report. You really can’t argue with it. The organisations truly winning aren’t just dabbling, they’re fundamentally redesigning their entire workflows with AI, and they’re unlocking massive, outsized value.

Proof, not promises

At Nearform, we pride ourselves on being on the front lines of innovation versus in the audience. So, what does AI-native engineering look like in practice?

• For a global pharma client, we used AI-native engineering to slash discovery timelines from six weeks to just two. Their AWS infrastructure was typically a multi-week effort. We had it up and running in minutes.
• We delivered a complex legacy platform for a major consultancy in eight weeks — four times faster than competitors’ estimates. That’s not just speed, it’s a totally transformed workflow, reducing manual report creation from days to same-day delivery.
• For a financial services firm, we built a complex data platform in just under four weeks, delivering twice the functionality of traditional approaches. We don’t just have a need for speed — we actually solve problems that others thought were impossible.

The new role of the engineer

In the AI-native era, the engineer’s role transforms from traditional coding to one akin of a conductor, directing networks of AI agents and systems. They’re right at the centre — the intent architect that translates messy business challenges into the precise, machine-readable directives that drive real outcomes.

This will see the rise of the ‘context engineer’, where the role isn’t just about providing hands — the value is now in orchestrating intelligence. CTOs need to keep an eager eye out for engineers and devs with strong soft skills — one that can really get to the bottom of what the business problem is, in order to find the right solution at speed.

Want to stay afloat?

Agile was a game changer. But as a differentiator, it’s dead. It promised agility, velocity and a new way to build software, and it delivered. But now it’s just the baseline, bread and butter for 94% of organisations. Agile already keeps organisations moving in the right direction — that was its entire purpose. The problem is that everyone now moves at roughly the same pace. Agile has become table stakes, not a competitive advantage. The true differentiator of this era is AI-native engineering, which introduces a fundamentally faster, compounding and machine-driven velocity curve.

Embracing AI-native engineering is the leading indicator of whether an organisation lags or leads in this new era. And just like the first sparks of Agile, there’s a familiar charge in the air… that feeling that the ground is shifting beneath our feet, that the rules are being rewritten, that possibility is expanding faster than most organisations can comprehend.

The question isn’t if AI-native engineering will redefine software delivery, it’s whether your organisation will step into this moment with intent — leading, experimenting, shaping the future?

Eager to be ahead of the game and learn where your organisation falls on the AI maturity spectrum? Get in touch: https://nearform.com/

Nearform partners with the world’s leading brands in complex or regulated industries, including financial services, healthcare, life sciences and the public sector.

In this conversation, Ciaran unpacks how AI can move beyond hype to drive real results by boosting profitability and security without sacrificing an excellent user experience.

Q: What’s the core challenge Nearform solves for leading organisations?

A: We’re laser-focused on business outcomes. It’s not about building tech for the sake of it. In complex, regulated industries trust and resilience will ultimately decide if AI goes live. Our edge is delivering AI and modern software systems that are fast, secure, transparent and tied to what really counts.

Our experts cut through the noise, navigating the tough stuff like governance and security that are needed to get beyond the proof of concept stage. For example, whether it’s automating Anti-Money Laundering (AML), personalising wealth management experiences, or modernising legacy systems to unlock customer data, we move quickly and securely to deliver real value in live systems, skipping those marathon transformation programmes that go nowhere.

Q: What’s the best approach for an enterprise to unlock AI? What does that look like on the ground?

A: The truth is, the best approach isn’t starting with AI, it’s starting with the business.

For example, so many banks have been told the myth, “fix your data, and one day, you’ll unlock AI”. But that’s not reality — it’s a recipe for endless spend, with zero outcomes — as by the time platforms are ‘ready’, priorities will inevitably have moved on and the business will be fed up.

That’s why we flip the script on introducing AI to critical workflows:

• Start with one value stream, ditching the platform-first mindset. We pick a real business process, like loan approval, risk, prospecting, etc, and take one thin slice end-to-end, customer to back office. The goal is to prove value (and build organisational confidence) in a focused area first, then scale from there
• Then go deep, not wide. We’ll dive in on one or two priorities, concentrating effort, rather than spreading it across endless pilots.
• For each slice, it’s about using the data you need and nothing more. You learn by doing in production, then scale what works.

This approach means creating cross-functional pods, to bring together experts across business, data, tech, risk and compliance functions. You want to create one blended team, focused on one customer outcome — think ‘mortgage approval’ or ‘pension onboarding’, owned end-to-end. That’s how we create speed, accountability and trust at the same time.

Q: Nearform works with clients across a number of regulated industries, such as banking and financial services. Can you share an example of what impact that work delivers?

A: Our work in financial services clusters in three areas:

Risk and remediation

In banking and insurance, we help accelerate massive AML and risk programmes. What used to need armies of people (scanning docs, closing KYC gaps) now gets done by AI, flagging only the real edge cases for humans. This massively reduces cost and time for each process, while elevating accuracy and audit trails. Just as importantly, it helps organisations scale these processes with stronger controls and clearer governance.

Wealth and asset management

We also help banks better serve high-net-worth individuals and different sectors like the emerging wealthy or teen savers. Banks have rich data on spending and life goals for each of their clients, and AI can help. It can surface more relevant next-best actions and prompt advisors to engage at the right moment — whether they’re about to start a family, buy a house or are learning about financial literacy and savings. Think fintech-like experiences, which are backed by the trust of a legacy institution.

Unlocking legacy data

Many banks are sitting on gold — decades of data — that’s locked away. Modernising used to be ‘too slow’ or ‘too risky’, but we can now upgrade or wrap legacy apps to turn that hidden data into AI-ready products, fast. That can deliver true ROI in just 12 months.

Although we keep client specifics confidential, these three challenges are prevalent and well-documented throughout the financial industry.

Q: What’s the number one misstep leaders are making in AI strategy?

A: We call it ‘the field of dreams’. That is, believing that if you spend years building a perfect data foundation, value will just ‘appear’. That’s not reality — in fact, all it creates is burned budgets and lost time, and by the time platforms launch, the goalposts will have moved.

Instead, leaders need to start with real business outcomes (like loan approval or fraud) and work backwards to the data they need. If investment isn’t mapped to revenue, cost or risk, it’s being wasted.

Q: With endless options, where should CTOs or CIOs focus their AI bets?

A: Pick one high-value workstream and go ‘all in’. Choose a commercial process, whether that’s mortgage approvals or credit decisions, and build an expert squad that’s obsessed with delivering results.

True leaders create focused pods that truly understand the data, context and compliance, inside out. That kind of deep focus brings speed and trust, and when you see the results, you can scale across adjacent workflows. Essentially, it’s about being excellent somewhere, not average everywhere. In practice, that means a small, senior, cross-functional team, working on clearly defined outcomes.

Q: What are the AI blind spots most executives fail to see?

A: Leaders often think they need to spend a fortune before they’ll see real gains, so they get sold on massive projects, like rebuilding global data platforms, with no real tieback to outcomes. Even worse, they miss how quickly AI can scale from pilot to production when trust and security are baked in.

Tech isn’t the real blocker, it’s the operating and delivery model: architecture, governance, engineering discipline, talent,and reporting lines. Leaders need to focus on unlocking existing value first, reviewing their ways of working and shift away from building for the sake of building.

Q: What tough decisions should leaders stop avoiding?

A: The hardest call is reorganising around outcomes that are built into products, not functions. AI works best when teams are cross-functional and centred around a specific product that delivers real value. This means upending traditional structures to ensure business, data and compliance are able to work as one united team.

That can be an uncomfortable leap, it’s needed to ensure that the AI journey is not throttled by structure or overhead, and can fully realise its potential to deliver meaningful value.

You can find out more by visiting Nearform’s website: https://nearform.com/

No one person or organisation has all the answers, and in such an environment, transparency and collaboration matter far more than polish.

There’s a certain kind of energy you feel when you go to speak on a panel and immediately sense that people are hungry for the conversation you’re about to have.

That was the atmosphere at Ireland House during SXSW — there were hundreds of engineers, designers, founders and business leaders from all around the world, ready to dive straight into the messy, exhilarating reality of AI adoption.

I hadn’t been to SXSW since 2019, before anyone had even heard the words “COVID”, “GPT” or even “aura farming.” Since then, SXSW has been through its own post-pandemic recovery, people outside of the tech bubble began using “Chat” in their everyday lives, and the majority of my LinkedIn feed now seems to be offering various forms of AI expertise.

That’s why, when walking up to the stage to talk about the reality of AI getting to production, instead of the hype, what struck me most was the level of engagement. There was such a mix of people in the audience, from all around the world, but there was leaning forward, note-taking and silent nodding that I haven’t really experienced since those years ago. After our on-stage conversation, I couldn’t believe the number of engineers from so many countries who came up to share a version of the same sentiment: “We’re living this too”.

Some referenced their own shift from pyramid-shaped teams to a diamond model, others were relieved to hear someone say out loud that senior engineers aren’t going anywhere and, in fact, are becoming more important as AI accelerates the base layers of delivery. Honestly, it felt less like a panel and more like a collective exhale. It was a moment of shared understanding in an industry that’s trying to build something entirely new. Yes, the ground is shifting beneath our feet, but that just means there’s uncharted land that we’re all discovering, in real-time.

For me, that’s the most exciting part of this moment in AI. It feels like the wild west. It’s messy and full of unknowns, but everyone’s learning together. No one person or organisation has all the answers, and in such an environment, transparency and collaboration matter far more than polish. It reminds me of why I love Nearform’s open source-influenced culture: the idea that learning in public, sharing breakthroughs and failures with equal enthusiasm, is what pushes an entire ecosystem forward.

We’re all learning as we go

A story that stuck with me was told by a fellow panelist from FedEx, who talked about how any corporate employee, technical or not, can now build internal AI agents using their in-house, no-code library. This is an organisation that “touches 99% of the world’s GDP” she pointed out, so the margin for error is minimal. Yet she was candid about the entire organisation learning as they go, openly admitting missteps and iterating along the way. That humility paired with ambition — and a laser focus on end user needs — is exactly the mindset enterprises need.

Now, most organisations don’t struggle with AI because of bad tools, but because they try to adopt AI as if it were just another tool. They may drop an assistant onto someone’s desktop and assume transformation will happen. But genuine impact requires much more, as we need to rethink the operating model, how individuals are incentivised and how teams are structured, how data pipelines are connected and ultimately architected and governed. It’s not the AI alone, but rather the ‘human plus AI interaction’, that needs to be carefully designed.

So, how is the rest of the world approaching AI?

One interesting theme that came up quite a bit at SXSW was the difference in how various regions are approaching AI. The contrast between the US and Europe was impossible to ignore. Many in the room expressed the view that American enterprises are moving faster, experimenting more boldly and integrating AI deeper into everyday workflows.

But Europe has something the US doesn’t yet: structure. More guardrails, clearer regulatory momentum, and even a push towards mandated AI literacy. It’s ironic, in a way: the US seems to be ‘winning’ on speed, but the EU may be building a stronger foundation for the long-term.

Testing in the age of uncertainty

Another lesson from the event was that traditional approaches to software testing simply don’t hold up anymore. We’ve spent decades building engineering cultures around deterministic systems and environments where the same input produces the same output, and quality can be validated with binary pass/fail checks. But AI breaks that paradigm.

Large language models and agentic systems are non-deterministic, and even with temperature turned to zero, the model can produce different phrasings, tool call sequences or reasoning paths on different runs. That means familiar safety nets — unit tests, integration tests, UI automation — are no longer sufficient. Instead, it requires an evolution in the entire discipline of testing. This may be one of the most underestimated shifts organisations face, especially those that have faced challenges in making the leap from AI in POC to AI in production.

Indeed, to build AI systems that behave reliably at scale, teams need evaluation frameworks, not just test suites. They also need expanded edge case testing — far beyond what they’re used to with deterministic systems.

A new beginning may mean saying goodbye, too

When a discipline reinvents itself as quickly as AI has for all of digital technology, it inevitably raises questions about what it means for engineering roles, and all of this change makes many people nervous that some may lose their jobs if they don’t adapt and evolve.

But I (as did a lot of the room), pushed back on the myth that ‘AI will replace engineers’. If anything, AI is making senior engineers even more important. As models take on more of the mechanical coding, humans become like orchestra conductors — managing agents and people alike, determining specs, identifying patterns, flagging risks and taking a system-level perspective for not only productivity but also safety.

I left SXSW feeling more optimistic than ever (despite my sore feet and hoarse voice). Not because the challenges surrounding AI aren’t real — they’re enormous — but because the willingness to tackle them collectively is growing, and fast. Companies are beginning to accept that success won’t come from chasing the latest model upgrade or vibe coding yet another demo.

It will come from designing systems that are observable, trustworthy and integrated into real workflows, from empowering teams rather than overwhelming them, and from adopting AI, not as a new way of building. And above all, from understanding that no one (yes, no one), has AI fully figured out. Which means everyone is still invited to help shape what comes next. And it can be anyone — from any country, any background — who has the ingenuity and creativity to find new solutions for these nascent challenges.

If this is the wild west, then SXSW proved one thing: we’re all playing pioneer together.

Nearform attended SXSW 2026 with Enterprise Ireland at Ireland House.

Peri Kadaster, Chief Communications Officer at Nearform, spoke on the ‘From Hype to Hard Problems: Building AI That Actually Scales’ panel discussion, alongside Kimble Jenkins, CEO, Orthosouth; Andrew Conolly, Founder, Wrksense; and Stephanie Cannon, VP, Digital Global IT, FedEx.

Peri also spoke on The Next Innovation podcast recorded live from SXSW, alongside Brenda Jordan, CEO, AskSobi; hosted by journalist Jennifer Strong. The episode explored “Do humans still matter in the evolution of AI?” — you can listen to this episode here.

https://nearform.com/

AI is reshaping the fabric of software engineering. However, while the first wave of adoption brought big promises, it also exposed crafts — ineffective tools, unsustainable practices, and a gap between the tech’s hype and measurable impact. At Nearform, we set out to change that trajectory and shape what comes next.

It won’t be long until every system is born AI-native. Purposeful, intelligent and engineered with progress at its core. According to a recent McKinsey survey, the majority of organisations that have embedded AI into their workflows have seen it improve innovation, with nearly half also reporting improvements in customer satisfaction and competitive differentiation.

This is a blueprint for enduring transformation. As pioneers in AI-native engineering (AINE), Nearform continues its 15-year tradition of pushing the boundaries of software development. Our teams and enterprise partners are leading this next wave of digital evolution, setting new standards in how software is conceived, built and trusted at scale.

A more intelligent way to build

AI succeeds when it’s foundational. Countless enterprises saw initial AI tooling fall short because it failed to address the core of the development lifecycle. Nearform’s approach is different from end to end — weaving AI into every stage of the software development cycle to bring precision and speed to the entire process.

Software demands more than code, it also needs intent, precision and trust behind that code. That’s where spec-driven development (SDD) comes into play.

At the core of durable AI-native systems sits a solid specification (or series of specifications). With SDD, product requirements, business logic and architectural requirements are all codified into a series of specifications, before a single line of code is written. These specs then become the core truths between human and AI, guiding agents to build, test and deploy with precision and consistency. This ensures clear, enforceable outcomes and removes ambiguity from the process entirely.

The AI-native engineering stack: engineered for performance

An AI-native engineering stack is a layered system that’s designed for enterprise scale. Each component of the stack magnifies the effect of the next, removing friction and automating processes from end-to-end:

● The spec defines the ‘what’ and the ‘why’, creating a clear blueprint.

● The AI agents interpret the spec to generate, review and test code.

● Automated enforcement guarantees that testing, security and quality standards are met.

This represents a bold step forward for ambitious enterprises, helping to unlock new sources of value from developing scalable, maintainable and safe software.

From talk to tangible results

AI-native engineering is more than just a concept, it’s a measurable improvement to the development process. By applying AI-native engineering, we focus on metrics that matter most to our enterprise clients:

● Higher acceptance rates for agent-generated code. In fact, our teams also cite higher delivery output, with 2x more functionality compared to traditional approaches.

● Faster merge times and accelerated delivery cycles. In the financial services industry, we recently delivered a new production system in less than four weeks, enabling early customer onboarding.

● Lower defect rates and more resilient systems. The early detection of conflicting requirements means a reduction in any late-stage surprises.

Our client data points to sustainable improvements of 10–15% in productivity, onboarding and prototyping — figures that aren’t only measurable, but also repeatable and compounding. In fact, well-suited greenfield projects with SDD are typically delivered 4x faster, and at the frontier, we can push to deliver 10x faster.

We’re redefining the meaning of progress for enterprise engineering, opening new horizons for what organisations can achieve. We’re visionary about what’s possible and relentless about what works, and everything we do is centred around building trust, scale and value into every layer of your business.

Ready to engineer the future with us? Visit www.nearform.com for more information.