AI in Business: What is really happening?

Delve into the rapid evolution of artificial intelligence with John Welsh, co-founder of True Worth AI.

John discusses the transition from AI as a question-and-answer tool to autonomous agent systems capable of economically viable work. From a consumer-facing novelty into an enterprise transformation force.

The conversation addresses governance, data quality, regulatory developments, enterprise readiness, and the redesign of operating models around AI-enabled workflows.

The discussion highlights the shift towards small, specialist models, the emergence of compound agent systems, and the growing need for AI literacy and operational oversight. It also examines the broader implications for the future of work, digital commerce, and the structure of the internet itself, positioning 2026 as a critical year in moving from experimentation to demonstrable business value.

Find out more about True Worth -> Here.

Key Take Aways

AI has moved from experimental R&D environments into mainstream enterprise usage, reaching mass adoption at unprecedented speed.
The market is at an inflection point: shifting from proof of concept to proof of value, with 2026 positioned as a defining year for enterprise ROI.
Agents—AI models with autonomy and tool-use capabilities—represent the next evolution beyond simple question-and-answer systems.
AI should be treated like a junior team member: requiring training, supervision, structured access, and quality control.
Poor data quality compounds risk in multi-agent systems, amplifying errors as tasks are handed off between models.
Governance, guardrails, and observability are critical as AI systems gain reasoning and decision-making capabilities.
Traditional multi-year transformation programmes are ill-suited to AI’s pace of change; iterative experimentation is required.
Small, specialist language models offer enterprises greater control, data security, and targeted performance than large frontier models.
Regulatory pressure is rising, but organisational self-governance must go beyond legal compliance.
Organisational AI readiness remains low despite high consumer-level adoption.
The future role of knowledge workers may shift from task execution to managing teams of AI agents.
The interface to the internet is likely to shift from human browsing to agent-driven interactions, fundamentally altering digital business models.

Innovatation

Deployment of autonomous AI agents capable of tool usage and end-to-end workflow execution.
Compound AI systems, where specialist agents collaborate sequentially across processes.
Use of secure AI sandboxes with synthesised data to enable safe experimentation.
Integration of small language models inside enterprise environments to retain data sovereignty.
Hybrid architectures combining local processing of sensitive data with external model calls for specialised tasks.
Observability and traceability layers to audit AI reasoning and decision pathways.
Agent-to-agent protocols enabling machine-mediated internet commerce.
Optimisation of digital content for AI agents rather than traditional SEO-driven human browsing.

Key Statistics

AI tools have grown to a couple of billion users, expanding many times faster than the internet.
Approximately two thirds of people globally have used some form of AI.
Only around one fifth of users report using AI extensively.
Hugging Face hosts over 2 million AI models.
Some organisations assume productivity gains of up to 40% from blanket AI licensing, though this assumption is challenged in the discussion.

Key Discussion Points

The speed of AI development is exponential and materially faster than previous technology waves.
AI is evolving from an “answer engine” to a system capable of economically viable work.
Enterprise implementation requires redesigning processes around outcomes rather than automating existing workflows.
Data quality and governance are foundational; poor inputs create compounding downstream failures.
Shadow AI within organisations poses governance and security risks.
Regulatory frameworks will shape risk management expectations.
AI governance must balance control with creativity to avoid stifling innovation.
Low-tech controls remain powerful mitigants against AI-enabled fraud.
Organisational literacy gaps exist both in AI capabilities and operational deployment.
Multimodal models (text, image, video, audio) are advancing rapidly and may disrupt creative industries.
Generic AI-generated content risks commoditisation, potentially increasing the premium on high-quality human creativity.
The internet may evolve into an agent-mediated infrastructure layer rather than a human-browsed environment.