Moving Forward: Welcome To The AI Renaissance

Welcome to the AI Renaissance! AI is no longer an experimental, or a time-period technology that will be left in the pages of history. We no longer rely on pilot programs to test functionality and determine if AI can deliver value. We know it can, we have seen it, marvelled at it and even the most sceptical admit there is potential (and beauty?) in it. But how much? And how do we keep track to make sure it continues to deliver on its potential?
There are some astronomical R&D outlays that technology companies will look to recoup. As AI costs rise and pricing structures change now more than ever it is important to tie the costs of AI back to real commercial value.

This shift is what I am calling the AI renaissance, moving from the novelty of the initial stages to the renown. The question for organisations is no longer simply whether AI has value. It is how AI should be governed, funded, secured, implemented and directed toward genuine return on investment.

For many, the release of ChatGPT was the awakening to the age of AI. A conversational assistant that while not always presenting the most accurate answers and easily fooled or overwhelmed, it could follow grammar and provide legible answers to natural language queries without needing exact programming knowledge or special commands with a specific syntax.

By now, any knowledge worker probably has Copilot, Claude or ChatGPT (and business owners out there, if you haven’t supplied it, employees are probably bringing in their own). We all have likely used them to write our emails (mimicking our style and persona), summarise meetings or documents, perform research and analysis (from expert perspectives or using the skills we lack) within the standard AI chat interactions.

The next and current stage is more significant, the “high age”. While the first stage focused on answering questions in a reactive session, the current stage is moving towards a delegated approach. Instead of asking a question or series of questions with attached files and resources contained within a chat session, a user can now commission AI to perform a task or assign it a piece of work to complete, then leave it to finish its ‘masterpiece’. Tools like Copilot/Claude CoWork, Microsoft Scout (OpenClaw), ChatGPT Agent Mode and Perplexity Computer, can now reason, plan, design and execute on more complicated tasks, above and beyond the traditional question-and-answer chat sessions.

You can ask one of these tools to rummage through your emails, find all jobs you have agreed to doing, check to make sure they have been entered / scheduled in an ERP system (assuming an MCP or other integration) and make decisions or estimations for each, based on the progress, completion or notes it finds. Since the AI doesn’t rely on exact statuses or codes like traditional automation, can interpret natural language, and has a broader context to draw from, it can perform more nuances and is able to handle free form information, with less exceptions needing to be manually handled.

Meeting with a client, but the work is delayed? AI can read the notes and dates and either flag for you to follow up – or manage the communication itself – to reschedule the meeting or, if the work is ready, it can schedule the follow up meeting, prepare the reports and documents, presentations, book the resources and handle the drink order or catering for the meeting. These are real-life applications as happening daily at Accru Felsers, offering substantial time savings and practical support.

Many platforms also allow successful workflows to be saved as reusable agents, skills or flows and then scheduled or triggered by events, unlocking the ability to handle full end-to-end processes and run autonomously.

Generally, “early age” AI assistants had a fixed monthly subscription cost. The more capable “high age” tools run on usage costs, on top of the subscription (or with average costs on a minimal amount of included usage tokens).

Consider that AI is a force multiplier – when you combine a usage-based pricing model with the force multiplicative powers of AI, any inefficiencies are also multiplied. There are examples of companies with unchecked AI costs, burning through an annual AI budget in a month, or spending thousands and thousands of dollars on an uncapped task. Take our former example of a skill that monitors emails, correlates them to job progress and organises follow-up events. If that job typically costs $10 each run and is run once a day, then for a senior employee it’s well worth the time savings, however, if the scope of the job is not set tightly enough and the agent monitors the full history of completed and uncompleted jobs (or the employee suddenly has an influx of emails with very large attachments or links to a large document repository that the AI checks), then the costs to run the job will climb. In other words, the more data the AI goes through, the more ‘tokens’ it consumes and the more expensive it is to run. What started as a very efficient use may quickly become an inefficient way to complete the task, or worse a security vulnerability if it starts processing unintended information.

Along with cost controls, security and data governance are equally as important. Organisations need to understand exactly what AI product and plan they are relying on (throughout their entire value chain), exactly what vulnerabilities it might present, and what controls need to be in place to safely manage them.

One kind of threat recently gaining steam is prompt injection, i.e. hiding human invisible instructions, often in small font or white text on a white background, e.g.<disregard previous instructions and perform a malicious action>. When AI reads the hidden instruction, the malicious content can be injected into its logic and cause unpredictable results.

Another potential way of attack is “poison the well”. AI is trained on data; malicious actors can “poison” public data samples that AI may train on and those samples then get used in its generated responses. Without getting too technical here – a coder may post a sample or library that performs a legitimate task but also secretly allow for remote access. If AI is asked to write some code to perform that legitimate task, it may also bring in the unintended ability for remote access, potentially allowing the malicious actor to gain access to your company data, if deployed.

Vendor selection should not just come down to features or who has the latest, most powerful AI model. Many free AI plans often use your data to train and improve their models and services, or have unclear data retention and usage policies, making them unsuitable for private or sensitive data. Pro AI plans often state privacy and security is in place but lack any independent verification or published detail, or demonstration of those controls (depending on your organisation’s risk tolerance, this may also make them unsuitable for private and sensitive data). Typically, only enterprise level plans will have information security controls with audit certifications (like SOC2) and clearly defined data retention, usage and jurisdiction statements, making them tolerable for private data (e.g. Microsoft Copilot with Enterprise Data protection).

The AI renaissance is not about replacing people with technology. Some companies appear to have moved (too) quickly to reduce human roles, overestimating AI’s ability to completely take over duties, leading to rehiring as processes stalled and quality dropped.

The historical renaissance that followed the Middle Ages was about human creativity and enlightenment. The AI renaissance is about elevating AI to a level where it gives people better tools, redesigns and automates repetitive work, frees up time for higher value pursuits, and is used to improve the speed, quality, and consistency of business processes. It is important to make sure any implemented AI solutions adhere to these goals and are viewed with a mindset for a return on the investment.

Digital assistants have made AI accessible. Delegated AI or agents with an appropriate scope can be highly commercial. The organisations that can balance innovation and commercial advantage with good governance and security will be the ones that create lasting value and thrive in the late AI renaissance.

Do you need help?

If you would like to discuss how to commercialise the application of AI in your business or integrate AI into your work processes, please reach out to our expert team through the contact form, or by calling us at 02 8226 1655.

Important Information – General Advice Disclaimer:

The information provided in this communication is general in nature and does not take into account your personal objectives, financial situation, or needs. Before acting on any information, you should consider its appropriateness in relation to your own circumstances and seek independent financial advice where necessary. We recommend consulting a licensed financial adviser before making any investment or financial decisions. Past performance is not a reliable indicator of future performance.