Microsoft Build 2024 : The future of work is now.

Microsoft wants to be your chosen Copilot for work. The developers conference this year marked a massive leap into a workplace where efficiency is key, intelligence is your companion and everyone benefits. And many companies are investing in this promise with several larger organisations setting targets for 90% of their workforce to be trained on Azure AI and the utilisation of Copilots, by the end of this year!

A large part of the allure behind adopting an enterprise solution such as Microsoft 365 with Azure AI is the promise that you can securely engage with your company's content in the cloud - based on a recent report by Microsoft and LinkedIn, many of us are using generative AI to help with a work related task without telling our employers. Having everyone on the same platform protects against the uploading of confidential information via a prompt to any of the free to access LLMs available.

A typical project with Copilots

Retrieval-Augmented Generation or RAG for short, is a fundamental element so let me offer an example to help explain how this works : you as the project lead will create a new project folder and upload all the documents related to that client and project, the moment these are uploaded they are being 'read' and broken down into chunks. Through the process of vector embedding the information is mapped and all similarities of context and meaning of words, segments, sections, titled files etc accumulate on a VectorDB (Vector database). By breaking down your information and assigning everything a vector value, the Copilot (through an LLM) shall retrieve only the information it feels is necessary based on your instruction. This not only greatly improves accuracy but also saves hugely on usage - making continuous API calls and running these models requires lots of processing power, finding efficiencies by only recalling what is required is incredibly important.

Once you have established your specific VectorDB, your team will engage with it by creating custom Copilots (Copilot Studio) based on their role in the project. In addition, ongoing contributions back into the 'folder' will expand the VectorDB. Microsoft is promising to offer you a complete solution architecture where development (GitHub Copilot), lives alongside real time data analysis using PowerBI (Fabric Real Time Intelligence) and product management all through their Copilot stack.

Trust and the human factor

Furthermore you can now invite Copilot into your Teams calls -  the Teams Copilot can listen in on the meeting to gather notes or answer any questions, drawing from your specific project data. This however does open up questions around trust and human engagement - soft skills shall remain an essential part of client conversation and management. What about the information? Financial institutions already have established security silos for communications across borders and along with Governments, I do not foresee this level of enterprise AI acceptance, for some time to come.

Efficiency at what cost?

Licensing for every user in your organisation is not cheap - Microsoft Copilot for Microsoft 365 at £24 | $30 p/user p/mnth - this excludes any additional Azure AI Cloud investment. The cost at harnessing all these fabulous capabilities has been promised to come down over time - but the latest will still be out of reach for many. Even if there are deals to be made, I see most SME (small to medium sized businesses) and institutions within research and education, simply unable to take advantage of what Microsoft are offering. Education in particular will see widening gaps of AI access - Khan Academy and Microsoft have however announced a partnership with the goal to bring AI to every classroom, for free (US only for now) - through clever utilisation of SLM (small language models) requiring less processing and lower token usage. This may or may not be linked to President Biden's AI initiative but without question the country that invests the most, will be leaping ahead of the rest.

Oversight but for how much longer?

During numerous demonstrations over the three day event "...no coding required..." was a recurring feature, often greeted with whoops and applause from the audience. No doubt there are efficiencies to be found to reduce the need for additional developers to conduct tasks such as code reviews or debugging - Microsoft is telling you that tasks that previously took hours can now be actioned in minutes. The irony that very smart people are working hard on contributing to their own future redundancy - is not lost on many that attended.  Only a handful of developers are required to validate any of the code generated by these latest models. But what if even this human step is removed? Very soon these Frontier Models will generate a more efficient coding language that will be indecipherable and make any degree of human oversight almost impossible.

Efficiency will speed up innovation. I foresee a rapid increase in breakthroughs in the field of medicine, as one such example. While disruption is an inevitable by-product of progress, I feel we are being driven largely by a sense of FOMO and at a pace that is leaving many behind.

Sources:

Microsoft Build 2024 - Microsoft Build home page, May 2024

OneLake the OneDrive for data - Microsoft, 2024

AI at Work is here - Microsoft and LinkedIn report, May 2024

GitHub Copilot Workspace - Microsoft github blog, 2024

Fabric Real Time Intelligence - Microsoft Learn, 2024

Microsoft Copilot Studio - Microsoft, 2024

Teams Copilot: getting started - Microsoft, 2024

Khan Academy and Microsoft partner to expand access to AI - Microsoft and Khan Academy, May 2024

May 2024

Written by Sean Simone, IOKA

← Features