We gathered founders, engineers, product leaders, and the Head of Innovation from the world’s fastest-growing software company, Lovable. Packed round a breakfast table, 40 storeys above central London, and sharing the realisation; software development isn’t just evolving. It’s being rebuilt from scratch. Here are the key insights that surfaced from the discussion.
We’re early (much earlier than the headlines suggest)
Despite how fast everything seems to be moving, the people closest to the work feel the opposite: we’re still at the very beginning.
Think YouTube’s early days: inconsistent quality, with rare flashes of brilliance. Those first videos were objectively terrible (but hinting at something transformative), just as today’s AI tools are impressive in flashes, and wholly unreliable in others. The potential was always there; people just didn’t know what to do with it.
The same goes for today’s models; for the most part they’re not limiting what organisations can or can’t do, it’s the systems surrounding them. Teams are being held back by workflows that were never designed for generative behaviour, and by organisational processes that assume clarity before exploration.
Truthfully, even if model development froze today, with no improvements, breakthroughs, or new releases, it would take years for companies to catch up to what’s already possible.
The gap between capability and adoption isn’t quarters; we’re talking in decades. So we’re not late. We’re not even behind. We’re pre-Netscape. Everything is still being invented, including how we think about invention itself.
The real competition isn’t another startup, it’s tomorrow’s version of your product
AI-native teams share a quiet assumption: everything they’ve built so far is provisional.
When you can rebuild a working feature in a few hours, you stop treating your current product as a fixed foundation. Instead of asking “How do we iterate on this?” teams ask, “Given what’s now possible, would we build this the same way today?” That question makes cannibalisation a strategy, not a risk.
It shifts attention from defending what exists, to actively challenging it. It makes roadmaps shorter, feedback cycles tighter, and strategy more fluid. And it creates a strange dynamic: the biggest threat isn’t a competitor, it’s the next version of your own product.
Deletion is becoming a strategy, not a failure
Deleting code used to feel like a step backwards. Lost effort, lost certainty and lost investment.
In AI-native teams, it’s now a sign of clarity. If AI can regenerate functional components in hours, preserving features that aren’t pulling their weight becomes more costly than replacing them. Features are treated like hypotheses: ship, observe, learn, and replace. The cost of standing still is exceeding the cost of starting again.
Enterprise teams, by contrast, are still wired for preservation. Budgets, reporting cycles, and internal politics are attached to every feature. Removing code feels risky, even reckless. Features linger long after their usefulness fades, creating drag that compounds over time.
The result is a widening gap: one world gains speed through deletion, the other accumulates friction through attachment. AI-native teams move faster because they embrace the impermanence of their work, while enterprises still cling to preservation.
Working software is becoming the new slide deck
The appetite for concept pitches is disappearing. People want to see something real, even if it’s rough, temporary, and destined to be rebuilt later.
Leaders described winning support and pitches with prototypes built over a weekend. Innovation teams are securing healthy budget, not with long documents or polished presentations, but with short demos that show what an idea actually feels like.
These prototypes may be completely rebuilt once they’ve proven the concept, but now the priority is speed, not perfection. Moving fast and making an idea tangible matters more than preserving the first version.
This shift is also broadening who can prototype. Marketers, operations leaders and analysts are building working concepts without ever touching a traditional design tool. Production-quality can wait; influence and insight can’t.
The biggest bottleneck isn’t model capability, it’s human clarity
Models have come a long way, but one of the biggest challenges of software remains: people don’t know what they want until they see it. They misdescribe goals, change direction when things become real, and struggle to articulate requirements in ways machines can actually act on.
But attempts to solve this with education have backfired. More tutorials, more handholding, more onboarding; they all seem to make things worse. Users simply switch off when they’re presented with abstract guidance. The key is understanding human attention: people need encouraging feedback within six or seven seconds, or their eyes glaze over and they move on.
The most successful AI products bake this reality into their design. They don’t try to extract perfect requirements upfront or guide users through lengthy explanations. Instead, they create environments where users can freely explore, try, fail, and iterate until their idea becomes visible.
Like learning to kick a ball; you don’t want ten hours of theory, you want to try, miss, and immediately try again.
Enterprises want AI transformation, but their foundations can’t support it
The conversation turned, inevitably, to the elephant in every boardroom: ambition is booming, but enterprise infrastructure is almost prehistoric.
Leaders get excited about agents, generative workflows and AI-powered processes, while their teams hit fragmented data, legacy systems, and governance frameworks built for a pre-AI world.
But there’s a deeper tension beneath this: enterprises want creativity, while also demanding determinism. The magic of modern models comes from openness, ambiguity, and probabilistic behaviour, while enterprises demand predictability, auditability, and tightly controlled outputs. Those two things can’t currently coexist.
The more a model is constrained, the safer it becomes – and the duller. The freer it is, the more creative, but the more unpredictable.
Until enterprises reconcile this contradiction, transformation will keep stalling between ambition and governance. Progress will require new processes, redesigned workflows, and a willingness to retire systems that can’t support this new way of working.
A new kind of builder is emerging, and they’ve never known a world without AI
Perhaps the most striking observation of the morning: some people building software today have never experienced a working world without AI. For them, AI isn’t a tool they use, it’s just the environment they operate in – like the internet for the rest of us.
These new builders orchestrate multiple models at once, work in parallel rather than sequentially, and treat complexity as background noise. They iterate instinctively, without the inherited caution of generations who feared breaking things or preserving every feature. They don’t “use AI”; they work inside it.
This sort of built in “instinct” is reshaping what teams look like. It’s why high-growth companies are hiring younger, earlier, and for range rather than resume. Experience still matters, but native AI intuition is becoming its own form of seniority.
Teams built around range are outpacing the teams built around roles
Another pattern was impossible to ignore: the old, well-defined roles are starting to crack. Businesses who want to scale reach while staying lean are building teams of generalists. Engineers who can design, designers who can build, PMs who can ship. And founders who can do all of that.
AI-native businesses are now looking for people who can take a problem from zero to done (enough) themselves, without waiting for a chain of specialists. Generalists aren’t “jacks of all trades” – they’re multipliers. They manage problems end-to-end, move faster, take bolder bets, and recover quicker.
A shift is happening, faster than organisations can comprehend
In the end, it was clear: we’re not witnessing a new phase of software, we’re watching the foundations get rebuilt in real time.
AI-native teams are working in a world where everything is provisional, deletion creates speed, and a weekend prototype carries more weight than the perfect pitch deck. They move quickly because they assume nothing is permanent.
Enterprises, meanwhile, are trying to plug new capabilities into old structures built for certainty and preservation. And a new generation of builders, people who’ve never known a world without AI, is widening that gap every day.
It’s messy. It’s energising. And it’s happening whether we’re ready for it or not.
More tables are coming
This GoodCore × Lovable roundtable was part of an ongoing series to strip away the noise and talk about what’s really happening in AI and software, not the press release version.
If you’d like to join a future session, drop us a message. We’re keeping the table small, but the conversations big.
