If you spend any time on LinkedIn or in tech news, you will have seen the prediction a hundred times: AI writes code now, so developers are finished. Agencies are finished. Anyone can build a website by describing it to a chatbot, so why would anyone pay professionals to do it?

We have been building headless websites, apps and software for over seven years, and our team uses AI tools every day. What we are seeing on the ground looks nothing like the end of development. It looks like the start of a period where developers can do far more than they ever could, where the quality of what gets shipped rises sharply, and where the companies that combine AI with real engineering skill pull away from the rest of their market.

This is not wishful thinking from an agency with something to protect. The data backs it up, and so does the entire history of our industry.

We have been here before

Every major leap in software has been announced as the death of the developer. High-level languages were supposed to remove the need for programmers, because you no longer had to write machine code. Open source was supposed to shrink the industry, because so much code became free. WordPress and Wix were supposed to end web agencies, because anyone could publish a site.

None of it happened. Every time the cost of producing software fell, demand for software grew faster. Economists call this Jevons paradox: when something useful gets cheaper, we do not use less of it, we use far more. Cheaper software production has always meant more software, more ambitious software, and more people employed to build it.

The current numbers suggest the same pattern is repeating. GitHub's 2025 Octoverse report counted more than 180 million developers on the platform, with 36 million joining in 2025 alone, over one new developer every second. That is the fastest growth GitHub has ever recorded, in the very year AI coding tools went mainstream. If AI were killing development, this would be a strange way to show it.

What the research actually says about AI and productivity

The honest picture is more interesting than either the hype or the doom.

Adoption is nearly universal. Stack Overflow's 2025 developer survey found 84 per cent of developers using or planning to use AI tools, up from 76 per cent the year before. Yet trust is strikingly low: only around 3 per cent of developers say they highly trust the output, and 66 per cent say their biggest frustration is AI solutions that are almost right, but not quite.

That phrase,almost right but not quite, is the whole story in miniature. Anyone who has worked with these tools recognises it. AI produces plausible code at remarkable speed. Whether that code is correct, secure, accessible and maintainable is a different question, and answering it requires exactly the skills experienced developers have.

The productivity research shows the same split. Faros AI analysed telemetry from over 10,000 developers and found that individuals using AI completed 21 per cent more tasks and merged nearly twice as many pull requests. But they also introduced 9 per cent more bugs, and at the company level there was no significant correlation between AI adoption and overall improvement. The gains were real for individuals, then evaporated somewhere between the developer's laptop and the business outcome, absorbed by bottlenecks in review, testing and deployment.

A widely discussed study by METR went further, finding that when experienced open-source developers used AI tools, they were actually 19 per cent slower overall. Even after finishing, they still believed the AI had made them around 20 per cent faster. This striking gap between perception and reality suggests the tools help most where developers lack context, but cost time where experienced judgement is needed to evaluate what the AI produces.

The lesson from all of this is not that AI does not work. It clearly does, and the tools are improving quickly. The lesson is that AI on its own does not produce better software. AI plus judgement does. The value has moved from typing the code to knowing what to build, how to structure it, and how to tell good output from bad.

Speed without skill just produces bad software faster

There is now a name for building software by prompting an AI and accepting whatever comes back: vibe coding. It is brilliant for prototypes and throwaway tools. In production, the results are turning into a case study in why skill still matters.

Veracode tested code generated by more than 100 large language models and found that around 45 per cent of samples contained vulnerabilities from the OWASP Top 10, the industry's list of the most common security risks. That figure stayed flat even as the models got better at passing coding benchmarks. Separately, researchers who analysed over two million AI-generated code samples found that roughly one in five contained references to packages that do not exist. Attackers have started registering these hallucinated names in public repositories and filling them with malware, a pattern that now has its own name: slopsquatting.

Meanwhile, research from Anthropic found that developers using AI assistance scored around 17 percentage points lower on comprehension tests than those who coded without it. The study found the gap was smaller for developers who stayed engaged with the reasoning, asking questions and requesting explanations, rather than simply accepting and submitting AI output. In other words, the people getting the most from AI are the ones treating it as a very fast junior colleague, not an oracle.

Pull all of this together and a clear picture emerges. AI removes the grunt work of development. It does not remove the thinking. Architecture, security, performance, accessibility, integration design and user experience remain human problems, and they were always the hard part. What AI does is free skilled developers to spend far more of their time on those problems, which is precisely where better websites and better software come from.

The bar is rising, because customers will not accept less

Here is the part of the story that gets far less attention than the job-loss headlines: what happens to user expectations when everyone can ship software faster?

They go up. Salesforce's Connected Customer research found that 88 per cent of customers consider the experience a company provides as important as its products or services. Zendesk found that 73 per cent of consumers will switch to a competitor after repeated bad experiences, and many will leave after just one. On the web specifically, speed still decides revenue: sites that load in one second convert visitors at roughly two and a half times the rate of sites that take five.

For years, plenty of businesses got away with mediocre digital experiences because building something better was expensive and slow. That excuse is disappearing. When your competitors can iterate in days rather than months, a clunky checkout, a slow page or an inaccessible form stops being a minor annoyance and becomes a reason customers leave.

This is the new normal we see forming. AI is not lowering the standard of what gets built. It is raising the floor, and it is raising customer expectations with it. The businesses treating their website as a box ticked years ago will find that box quietly reopened by everyone around them.

What the winners are doing differently

The companies pulling ahead are not the ones that replaced developers with AI, and not the ones that banned it. They are the ones that redesigned how they work so that AI and engineering skill multiply each other. In practice, that looks like this.

●They use AI where it is strongest:generating the repetitive scaffolding of a build, writing tests, migrating content and data, exploring three design directions in the time one used to take, and prototyping ideas that would once have died in a backlog because they were too expensive to try.

●They keep experienced engineers where they matter most:shaping the architecture, reviewing everything that ships, owning security and performance, and making the hundreds of small judgement calls that separate software that works from software that lasts.

●They reinvest the time AI saves rather than banking it:the hours no longer spent on boilerplate go into the things customers actually notice, such as faster pages, smoother journeys, better accessibility, richer integrations, more polished detail. The budget that used to buy an adequate website now buys an excellent one.

That last point is the real opportunity. For a given budget, the amount of quality you can afford has jumped. Ambitions that were previously enterprise-only, such as fully bespoke headless builds, deep personalisation and genuinely fast global experiences, are now within reach of mid-sized businesses and start-ups. The constraint is no longer raw development hours. It is the thought, planning and execution wrapped around them.

Empowered developers, better web

So no, we do not believe AI kills developers. The evidence points somewhere much more useful: AI is becoming the biggest amplifier of developer skill the industry has ever had. More developers are entering the field than ever before. The ones who learn to direct these tools well are shipping more, exploring more and polishing more than any team could a few years ago.

The gap that matters now is not between companies that use AI and companies that do not. Within a year or two, everyone will use it. The gap is between companies that pair AI with strong engineering thinking and those that use it to ship mediocrity faster. The first group will set a standard of digital experience the second cannot match, and customers will vote with their attention.

That is the future we are building for at Lemon Hive. Our teams in London, Oxford and Dhaka use AI throughout our headless web, app and software projects, always with senior engineers directing the work and owning the quality of every line that ships. The result is what this whole shift promises when it is done properly: better websites, better software and better experiences, delivered faster, for businesses that want to rise above their market rather than blend into it.

If you are wondering what that could look like for your business or your agency's clients, we would be happy to show you.