App development in 2026: code? What code?

Coding in terms of using a coding language has since given way to telling an AI model what you need, and then pushing that code into production (with mixed results).

Democratised coding 

Source: Lenovo. Bhatia.

Sumir Bhatia, President, Asia Pacific, Infrastructure Solutions Group, Lenovo, said that AI will place innovation into more hands in 2026. "Moving into 2026, perhaps the most exciting shift is seeing how AI changes who can participate in innovation – the democratisation of AI. Natural‑language interfaces and agentic AI allow domain experts, doctors, plant managers, supply‑chain leaders, to design and orchestrate AI‑driven workflows without needing to be AI specialists," he said.  

"What this means is that innovation will no longer be confined to data scientists or IT teams. With agentic AI, everyday professionals can orchestrate workflows, accelerating experimentation and time to value. Combined with secure, well‑governed infrastructure, this trend will reshape industries from healthcare to manufacturing." 

English-language coding

Gopi Duddi, CTO, Couchbase, believes that English will become the most popular coding language. "Programming will move toward natural conversation as people rely on English to instruct systems rather than learning specialised languages," he said. 

"This shift increases the number of people who can automate tasks and create small programs because the barrier to entry becomes far lower. As more people generate information in unstructured form, the volume of data needing storage and retrieval will rise quickly." 

Duddi forecasted that traditional programmers will uplevel their skills and guide the rewriting of complex systems while new creators will rely on simple conversational prompts. "The result is an environment where data platforms must store natural inputs and support a growing population of domain specialists: for example, a doctor could theoretically write their own program," he said.

No-code no-go

Ways to make coding easier have existed for a while, of course, low- and no-code methods among them.

"No code’s on its last legs – it’s being snuffed out by vibe coding. AI-driven development tools will be the final knell for no code as we know it, with its remit curtailed in this new coding landscape," said Raymond Kok, CEO, Mendix.

With no-code methods, code is created by dragging and dropping visual elements into software. Vibe coding may replace the actual impetus for the code, but the visual interface remains, Kok elaborated.

"Soon, it will be very clear that using programming as a means to express your intent for a software system is no longer efficient. Now that AI can do all the heavy lifting in the back, models will become a more dominant representation of software engineering," he added.

"So, instead of messing around with a bunch of code, they will let you work with visual models to actually express your intent. We’re already seeing this with some of the newer tools available that are focused on agentic workflows."

The cracks in vibe coding

Some feel that the Gartner's 'trough of disillusionment' has begun to occur for vibe coding. "The hype for gen AI and vibe coding will – if it hasn’t already – officially peak in 2026. The initial euphoria of lightning-fast app generation is dissolving as organisations face a sobering truth: those minute-made apps are ticking time bombs of technical debt and risk," said Hans de Visser, Chief Product Officer, Mendix.

"Vibe coding platforms are hemorrhaging users, with traffic nosediving (some suspect 30-40% losses), further emphasising the issue of code that nobody understands or truly owns. Maintenance has become the biggest stumbling block for AI-assisted coding IDEs – where initial productivity gains are quickly offset by the increased time required for changes and course correction." 

De Visser explained that current large language models (LLMs) are not dependable for coding, the believes it could take years before they achieve "compiler-like reliability for traditional code abstraction". "For coding, this means that developers cannot yet rely on natural language input to consistently yield secure, runnable code. Working within development platforms that enable AI while also enforcing existing security guardrails and checks will be paramount for creating reliable, consistent and workable outputs in 2026," he suggested.

"This approach ensures any generated output meets the standards of human-written code and can withstand industry scrutiny."

Low-code may survive

A form of low-code integrated with AI may fix the problem, said Menno Odijk, Field CTO, Mendix. "There are so many holes being created by AI production applications. So much so that startups are arriving to solve the problems created by AI-built applications… but the solution is low code. The simple reason is because low code exists as building blocks, and all those building blocks are approved, governed, and secure – AI can only build upon those existing and trusted building blocks," he pointed out.

"It’s vibe coding, just without the hangover. And we’re likely to see more organisations realise this in the near future." 

First vibe, now swarm

Another concept, swarm coding, is also emerging - getting multiple agents to work together to create usable code. The concept is seductive enough, but comes with unique caveats.

"While it’s really early days, there are some organisations thinking about how to apply agentic swarm coding to building applications. The challenge, however, is that if you have multiple AI agents working together across multiple technologies, they all need to have a common definition of the information being processed. They all need to use the same terminology, effectively. So even with this new iteration of coding, data will still be the biggest challenge for application development," Odijk said.

In addition to ensuring all the AI agents are on the same page before they begin, the margin of error adds up very quickly with a swarm, de Visser said. "We’re seeing attempts to do, for example, agent orchestration and multiagent systems without too much formal orchestration. So you leave it basically to the LLM to do all of the coordination, but the technology is really in its infancy," he emphasised.

"If an AI agent is reliable 80% of the time, you can course-correct to get a higher percentage of accuracy. But if you multiply this, e.g. with five interdependent agents the quality degrades to 30% accuracy, which is much more challenging to correct."

"It’s a big problem and no one’s tackled it yet. For swarm coding to be effective in 2026, there needs to be an improvement in predictable, accurate outcomes," de Visser stressed.

Governance

“Even beyond the earlier citizen development phase, AI has enabled more and more people to create more and more software. Making sure that the software being produced is of good quality is becoming increasingly difficult. This is where having proper governance in place at an organisational level comes into its own – we need to avoid creating even more shadow IT. Governance will be critical to that,” said Sakshi Dhakad, Head of Product Management, Mendix. 

Philip Miller, AI Strategist, Progress Software, framed the problem as a lack of end-to-end workflows. "Coding copilots are quickly becoming table stakes, but the next horizon is end-to-end dev workflows,
requirements synthesis, test generation, secure refactoring, and compliance evidence. GitHub’s
enterprise research with Accenture found significant time savings and satisfaction gains, and broader
industry reporting points to sustained productivity lifts as adoption deepens," he said.

"Expect SRE and data-engineering agents to join the party, tightening feedback loops between app code, data products, and infra. The differentiator isn’t 'who uses a copilot', but who wires it into governed repos, policies, and automation to improve quality, traceability, and security posture."

Dev refers to development. IDE stands for integrated development environment. SRE stands for site reliability engineering, while a repo is an abbreviation for 'repository'.

Explore

App development in 2026: how teams could evolve 

Hashtag: #2026Predictions