AI surge to bring chaos before control, warns CEO

OutSystems chief executive Woodson Martin has predicted that artificial intelligence will increase complexity across large organisations in the near term, even as it speeds up software creation. He said many companies risk unmanaged growth in AI-generated applications and agents as they move from experiments into production systems.

Martin based his outlook on conversations with more than 80 chief information officers from sectors such as banking, insurance and manufacturing. He said these organisations face rising pressure as they adapt generative AI and agent-based systems for core operations.

He argued that many AI forecasts underestimate the disruption for established enterprises. He said the biggest beneficiaries of the current AI wave will be large user organisations rather than technology vendors.

According to Martin, many CIOs now expect to build most of the applications and agents they need for themselves. He said this shift will reduce reliance on traditional packaged software.

Rising complexity

Martin said AI is already speeding up software development. He cited "vibe coding" techniques that turn work that once took weeks into tasks completed in minutes or seconds.

He warned that most organisations focus only on the build phase of software projects. He said this creates future bottlenecks in quality control, security, maintenance and updates.

Martin expects 2026 to mark a turning point. He said IT teams will focus more on containing and auditing ungoverned AI-generated applications and AI agents.

He said organisations that use AI for governance across their full software portfolio will gain the earliest advantages from AI-driven development.

Fragile AI agents

Martin said most AI agents will not work reliably in production environments. He contrasted impressive demonstration systems with the realities of live enterprise infrastructure.

He said production environments involve constantly changing APIs, messy or incomplete data, conflicting business rules and complex identity and permission models. He added that non-deterministic AI behaviour produces unpredictable outcomes.

Most autonomous agents will require tight orchestration layers and human oversight, according to Martin. He said this dependence will drive demand for new types of platforms. "Autonomy only works in fantasy. It's orchestration that wins in reality," said Martin, Chief Executive Officer, OutSystems.

Platform over model

Martin said the rush by companies to build their own large language models has slowed. He sees smaller and more specialised models gaining ground in enterprises.

He expects a small group of large models to dominate consumer use. He said enterprise leaders will instead select from vertical or small language models and connect agents to multiple models for different scenarios.

He said ownership of the model will matter less than ownership of the lifecycle. He expects platforms that manage secure, governed, multi-model agent orchestration across complex organisations to capture more value.

Shift in value

Martin warned that unmanaged AI creates significant operational and compliance risks. He cited hallucinations, policy breaches, data leaks, model drift and incorrect workflow generation as major concerns.

He said the ability to ensure correctness at scale will become more important than simple software generation. He expects the market to place a premium on integrity in AI-driven systems. He described a new mantra of "trust > velocity".

He forecast that "shadow AI" will surpass past problems with shadow IT. He said non-technical users can now create production-grade code and workflows with large language models without central oversight.

He described unapproved models and agents as an existential risk. He said an unvetted tool in the hands of a business user can generate code, create autonomous workflows or move sensitive enterprise data without controls.

Governance spend

Martin expects IT budgets to rise as organisations invest in AI oversight. He said any deflationary effects from automation and lower development costs will be offset by new spending.

He said companies are adding new security layers such as runtime defences and guardrails against prompt injection, data leakage and rogue agent behaviour. He also expects more model oversight, including continuous evaluation for performance degradation, model drift and bias.

Martin said organisations face new compliance obligations. These include alignment with emerging frameworks such as the NIST AI Risk Management Framework and preparation for specialised AI audits. He also expects a "desperate scramble" for skills in AI engineering and governance.

He argued that code itself is becoming cheap while architecture becomes expensive. He said AI can generate functional code, which reduces its strategic value. He expects value and cost to concentrate in system architecture, data modelling, integration and lifecycle governance.

Business model tests

Martin believes "agentic AI" will put more pressure on leaders to rethink business models. He expects the focus to move from efficiency gains towards new types of services and revenue structures based on automation at scale.

He said AI agents will allow faster experiments with new business models. He said leaders will be able to scale ideas that succeed and withdraw from those that fail with lower personal risk.

He predicts that companies in highly regulated sectors will embed compliance into AI implementations ahead of government mandates. He cited finance, healthcare and manufacturing firms that already operate under strict rules.

He expects these organisations to adopt model traceability, mandatory responsible AI audits, architectural checks and role-based access restrictions. He said this approach will allow them to give agents more responsibility in areas that affect people and infrastructure and reduce the risk of high-profile failures or loss of trust.

Developers in demand

Martin said enterprise developers will gain value, not lose it. He argued that AI can automate general coding tasks but cannot manage systemic complexity.

He expects the most skilled developers to become five times more productive. He said this trend counters the idea that AI will replace them.

He said top-tier engineers will move from writing routine code towards directing networks of AI agents. He expects these specialists to become harder to hire and more central within organisations. "As top-tier talent transitions from writing boilerplate code to conducting a symphony of AI agents, they will be harder to find, increasingly leveraged, and dramatically more valuable," said Martin.