RO-AR.com: AI in Business: What is really happening

Key Take Aways

1. AI has moved from a specialist technology into the everyday business lexicon, with tools such as ChatGPT, Gemini and Claude now widely recognised and used.
2. The pace of AI change is described as faster than any technology shift seen by the speaker in 25 years in technology.
3. AI is at an inflection point: it is moving from being an “answer engine” towards performing economically viable work.
4. 2026 is positioned as the year organisations shift from proof of concept to proof of value.
5. The agentic AI model is central to this next phase. Agents are described as AI models with autonomy and the ability to use tools.
6. Organisations should start with business outcomes and work backwards, rather than beginning with the technology itself.
7. AI should be viewed as a junior team member: useful and capable, but requiring training, direction, supervision, guardrails and review.
8. Data quality remains foundational. Poor data will undermine AI outcomes and can compound errors across multi-agent systems.
9. AI governance is underdeveloped in most organisations, despite the growing need for policies, controls, risk assessment and auditability.
10. Traditional 12- or 24-month transformation plans may no longer be suitable in the AI age, given the pace of technological change.
11. Small, specialised language models may become increasingly important, particularly where organisations need control, lower compute requirements and stronger data security.
12. AI will reshape work, but the greatest opportunity is to let humans focus on creativity, judgement, goal-setting and management while AI agents handle more routine or lower-value tasks.

Innovation

• Move from AI as a question-and-answer tool to AI as a mechanism for economically valuable work.
• Use agents as AI models with autonomy and tool access, effectively giving AI “a pair of hands” to perform tasks.
• Design processes around desired outcomes first, then identify where AI can create value across the workflow.
• Build compound systems made up of multiple specialist agents, each performing a specific function such as customer interaction, data collection, analysis or remediation.
• Treat AI agents as junior members of staff, with role definition, training, oversight, permissions and performance monitoring.
• Use sandboxes with synthetic data to allow employees to experiment safely with AI tools without exposing personal, corporate or sensitive data.
• Apply stage gates to move AI ideas from experimentation into controlled production, ensuring security, safety, reliability and governance.
• Introduce observability and traceability so organisations can monitor AI activity in real time and audit decisions afterwards.
• Use small language models trained on specific organisational data or use cases, rather than relying solely on large frontier models.
• Combine local processing for sensitive data with external models for non-sensitive tasks such as web search or deep research.
• Shift from search engine optimisation to optimisation for agents, ensuring content and services are accessible to AI systems rather than only human web users.
• Prepare for agentic commerce, where agents search, compare, transact and interact with other agents on behalf of users.

Key Statistics

• AI is described as having grown to a couple of billion users.
• The speaker states that AI has grown many times faster than the internet did.
• The speaker has been in and around technology for about 25 years.
• Around two thirds of people globally are described as having used some form of AI, such as ChatGPT, Gemini or Claude.
• AI was previously associated with large R&D departments, seven-, eight- or nine-figure budgets, and teams of PhD-level specialists.
• Hugging Face is described as having more than two million AI models available.
• Some organisations assume that giving employees ChatGPT business licences could make them 40% more productive, but the speaker challenges this as an oversimplification.
• The EU AI Act is highlighted as coming into force.
• AI-generated video and image models were described as having materially improved in 2025.
• The speaker predicts that in 2026 awards may need to consider AI-assisted or AI-created media.
• The speaker estimates that only about a fifth of people would self-confess to using AI a lot.
• The speaker predicts that the internet may change significantly within a five- to ten-year window, and possibly sooner.

Key Discussion Points

1. AI has become mainstream, but the business conversation needs to move beyond curiosity and experimentation.
2. The practical business shift is from proof of concept to proof of value.
3. Agents are the next major development because they can use tools, exercise autonomy and execute tasks.
4. Organisations need to ask “what if” questions and redesign processes around what is now possible.
5. AI should amplify people by removing work they do not want to do or that machines can perform effectively.
6. AI agents need to be managed like junior employees, with training, checking and clearly defined boundaries.
7. Shadow AI is already a risk in organisations, whether leadership is aware of it or not.
8. Good AI implementation requires strong design, architecture, guardrails and governance.
9. Bad data creates poor AI outcomes, and in compound agent systems errors can worsen as information moves between agents.
10. AI transformation still requires first principles: good processes, good data, proper instructions, checking and change management.
11. Governance needs to balance control with experimentation, avoiding both excessive constraint and uncontrolled deployment.
12. The future internet may become an enablement layer for agents rather than a direct browsing experience for humans.

Description

This podcast is a discussion between Chris and John Welch, co-founder of True Worth AI, on what is really happening with AI in business. The conversation focuses on the shift from AI as a widely discussed technology to AI as a practical business capability that can perform valuable work.

The podcast argues that organisations are entering a new phase in 2026, moving from proof of concept to proof of value. A central theme is agentic AI: AI systems that have autonomy and can use tools to perform tasks. These agents are compared to junior members of staff who need training, direction, supervision, permissions and review.

The discussion is particularly relevant for senior leaders because it highlights the organisational implications of AI adoption. These include data quality, process design, governance, security, observability, traceability and employee readiness. The speakers caution against treating AI as a silver bullet or assuming that enterprise-wide licence rollouts will automatically generate productivity gains.

The podcast also explores how the technology landscape may evolve, including the use of small specialist models, multimodal AI, AI-generated media, authentication challenges, agentic commerce and the changing role of the internet. The overall message is that AI will require organisations to rethink how work is designed, how people are trained, how controls are applied and how value is created.