AI Insights: The Journey to Becoming an Agentic-First Enterprise

Every few generations humanity experiences a milestone moment. From the Stone Age to the Renaissance. From New World exploration to the Industrial Revolution. And more recently the arrival of personal computers in the 1990s. Now, with the advent of AI, we’re here again.

Although we remain at a relatively early stage of AI evolution, we’ve already made a seismic step – from simple conversational tools to highly functional autonomous entities with the development of AI agents. To date, this has had the biggest business impact within technology departments through the evolution of code generation.

The proportion of computer code written by autonomous agents is increasing rapidly. OpenAI President Greg Brockman noted at a recent conference that agentic tools handle roughly 80% of code generation. Similarly, Anthropic disclosed that more than 80% of the code merged into its own production repositories is authored by Claude.

This is freeing engineers from repetitive coding to focus on tasks with more strategic and creative value, and greater business impact. Along the way, we’ve learning about the problems of not providing the right guardrails and governance. Because, without sufficient context and control, agents fail to complete required tasks or hallucinate, delivering wrong and sometimes dangerous outcomes.

Looking beyond engineering

Companies are hungry to capitalise on the AI boom, approaching software engineers to integrate it across their operations. Their underlying needs are typically split into three distinct, quantifiable categories:

• AI in the Software Development Life Cycle (SDLC): This is the easiest baseline task because engineering departments already possess mature testing, release, and governance best practices to wrap around coding agents to make them more stable and reduce the risk of hallucinations.
• AI Tools Beyond Engineering: The greatest business potential occurs when non-technical branches (marketing, growth, HR, data, finance) gain equal access to AI. Just as Microsoft Office standardised digital capabilities across all business tiers, future AI must universally elevate the output of every employee.

• Applied AI Within Products: Embedding AI natively into the core consumer or operational workflow of a business. This fundamentally requires the company to be built AI-first.

  Keeping AI as the sole domain of the technology department hinders categories 2 and 3. While roughly 75% of workers use AI right now, 61% refuse to let agents work autonomously due to a lack of trust and fear of hallucinations. This barrier will remain if engineering hoards advanced prompt and agent structures, potentially leaving non-technical departments lightyears behind, preventing companies from getting the full impact and value from AI.

Getting there with an ADP

So how can you enable non-technical people across your business to get the most from AI tools in their day-to-day roles and through automated workflows? The answer lies in an agentic developer platform (ADP) that standardises AI specifications and practices across an organisation.

Around 90% of businesses already have an internal developer platform (IDP)* created to streamline software infrastructure management and development. Shielding developers from the underlying complexities of cloud architecture, this approach has successfully cut software deployment cycles down from months to weeks. However, its benefits remain strictly within the engineering team.

Integrating a Model Context Protocol into an IDP, unlocks full natural language capabilities, creating an ADP. This means that instead of clicking predefined buttons or writing code, developers can interact with the infrastructure by simply conversing with it. The platform stops acting as a rigid set of scripts and becomes an intelligent, reasoning system that interprets the user's intent, creates the deployment path, and handles the execution autonomously.

The true breakthrough of an ADP, however, is not that it makes life easier for engineers, but that it democratises software creation for the entire enterprise. For example, an ADP enables an HR manager or a marketing analyst to safely build and deploy their own automated software systems using natural language, rather than coding. All that’s required is teaching employees how to think like engineers when it comes to prompting and managing AI agents.

Shifting to an engineering mindset

Because AI agents act as an amplifier of the inputs they receive, giving high-powered autonomous tools to teams without structured thinking will only scale operational chaos. So, the key to optimising AI across an organisation doesn’t only lie with an ADP, but also requires non-technical teams to adopt four engineering principles to build and deploy agentic workflows safely. These include:

1. Thinking in first principles and algorithms: To work with an ADP, employees must view their daily tasks as a sequence of logical steps, breaking down a business goal into its smallest, unarguable components. For example, a finance professional cannot simply tell an agent to “audit the monthly expenses”. They must map out the exact logic: “On the 30th of the month, pull the CSV from Source X. If a line item exceeds £5,000, flag it for approval. If the 30th falls on a Saturday, execute this on the preceding Friday.” The human writes the airtight spec; the AI acts as the runtime.
2. Taking a scientific approach to hallucination management: Non-technical users must adopt a scientific, iterative approach to problem-solving. Teams must treat an agent like it’s a brilliant, but completely stateless worker who is reborn fresh every session. If an agent makes a mistake, the user shouldn't abandon the tool. They must isolate the variable, tweak the prompt or data context, and test it again under controlled conditions.
3. Implementing multi-layer verification: Non-technical departments must learn to build verification loops into their autonomous workflows rather than leaving agents entirely unsupervised. For example, if a marketing team builds an agent to generate advertising copy, they should not manually read every single output, but instead design a multi-layered workflow: Agent A writes the copy, Agent B acts as a critic to evaluate it against compliance guidelines, and Agent C arbitrates any disagreements between them before final deployment.
4. Imposing strict knowledge and context guardrails: Non-technical users must become highly disciplined regarding data boundaries and context window limitations, designing strict scope and role-based access rules for their custom agents. For instance, an HR professional must structure an agent's memory so it only pulls data from approved, clean folders. They must understand that overloading an agent with a chaotic dumping ground of unorganised files will confuse its “long-term memory” and result in sub-standard output.

This mindset shift cannot happen in a vacuum. It is the core responsibility of the engineers to stop hoarding this structural way of thinking and instead institutionalise it, transferring these mental frameworks to the rest of the company so every department can build independently and safely. Only then can the full impact and value of AI be liberated.

Find out more about successfully integrating AI across your business by watching Ahmed Tokyo's full Mindera AI Insights 2026 presentation below…

*Source: Platform Engineering 2.0: An evolution for the AI era — https://platformengineering.org/blog/introducing-platform-engineering-2-0-an-evolution-for-the-ai-era