New generative AI tools have been developing rapidly and are now widely used, including by workers in doing their jobs. Microsoft worked with firms across industries to run a large field experiment to measure how access to one of these tools changes work patterns. The experiment ran during the early rollout of Microsoft's M365 Copilot (Copilot), a multi-part generative AI tool that integrates generative AI into components of Microsoft's M365 suite (including Word, PowerPoint, Outlook, and Teams). M365 Copilot is designed as a general purpose tool to help workers digest information by summarizing emails, meetings, or documents, create new content by drafting emails, documents, or slide decks, and retrieve information either from the web or across any company email, chat, or document to which the worker has access. We worked with firms to randomize access to Copilot and got permission to use several months of anonymized metadata on workers' behaviors in Outlook, Teams, and Office, both before and after access to Copilot.

While it's unclear if mainstream PC users are actually using Microsoft's Copilot AI, the company claims that businesses using MS 365 Copilot are seeing plenty of benefits. According to a Microsoft survey, Copilot users at Honeywell save up to 92 minutes per week, while customer service agents at Teladoc are saving up to five hours a week by using the AI tool to draft responses to questions. Now that we're a year beyond the MS 365 Copilot launch (at a costly 30 per seat), Microsoft is eager to throw more AI features at corporate drones. Most intriguingly, Microsoft is upgrading its Business Chat app, which so far has been a way to interact with Copilot's across your emails, calendar entries and other data, alongside data from your organization. Now it's getting better collaboration with the addition of Copilot Pages, which will serve as a sort of "multiplayer" way to share AI generated content with your coworkers.

To understand the impacts of AI-driven coding tools on engineers' workflow and work environment, we utilize the Jellyfish platform to analyze indicators of change. Key indicators are derived from Allocations, Coding Fraction vs. PR Fraction, Lifecycle Phases, Cycle Time, Jira ticket size, PR pickup time, PR comments, PR comment count, interactions, and coding languages. Significant changes were observed in coding time fractions among Copilot users, with an average decrease of 3% with individual decreases as large as 15%. Ticket sizes decreased by an average of 16% across four companies, accompanied by an 8% decrease in cycle times, whereas the control group showed no change. Additionally, the PR process evolved with Copilot usage, featuring longer and more comprehensive comments, despite the weekly number of PRs reviewed remaining constant. Not all hypothesized changes were observed across all participating companies. However, some companies experienced a decrease in PR pickup times by up to 33%, indicating reduced workflow bottlenecks, and one company experienced a shift of up to 17% of effort from maintenance and support work towards product growth initiatives. This study is the first to utilize data from more than one company and goes beyond simple productivity and satisfaction measures, considering real-world engineering settings instead. By doing so, we highlight that some companies seem to benefit more than others from the use of Copilot and that changes can be subtle when investigating aggregates rather than specific aspects of engineering work and workflows - something that will be further investigated in the future.

Microsoft just announced its Copilot AI chatbot is integrating with OpenAI's latest model, GPT-4 Turbo, and the image generator DALL-E 3, among other upgrades. This should drastically improve the overall functionality of the service, just in time for its one-year anniversary/birthday. Wait, do AI chatbots have birthdays? GPT-4 Turbo integration will allow Copilot users to tackle complex tasks that would cause previous iterations of the software to sputter into madness. The last generation allowed for just 50 pages of text as a data input, while GPT-4 Turbo accepts up to 300 pages. The integration is currently being tested by select users, with wider availability in the next few weeks.