Workplace AI just had a terrible week

But get ready to bluff your way in 'Agentic AI'

We’re going to witness a lot of Microsoft pieces over the next few months - the company is soon to hit its 50th birthday and still regularly finds itself at the top of the biggest companies charts. (Based on market cap as I write they are 3rd behind Apple and Nvidia but it pogos around). To mark the impending anniversary Wired published a long-read exploration of the company’s cultural reinvention under Satya Nadella that is strongly recommended.

While the champagne is chilling, Microsoft’s Copilot AI just had a horrible, horrible week. One of the pushbacks on LLM AI tools is that some products are being overhyped above their current level of competence. In the case of MS Copilot it’s gone beyond that into the product being straight-up untrustworthy. Firstly the word-of-mouth on the product hasn’t been great, Gartner published a report by IT leaders and less than 4% said Copilot currently offered significant value to their companies.

The boss of Salesforce, Marc Benioff went as far as to compare Copilot to Clippy, the justifiably maligned cartoon assistant of Windows 97, ‘Microsoft has really disappointed so many of our customers. They have not delivered any competitive capability.’ Benioff in turn promised the power of agentic AI to his own customers. (For more on this see below).

But the issues of Copilot go deeper than that. The magic of the product is that it draws from company internal data, searching it like the web, to populate its answers, in the form of emails, presentations or documents. But the problem has been that Copilot has been snooping around company files, often spilling secrets it wasn’t meant to share. Some users have found the salaries of colleagues, or sensitive HR files - something that Business Insider explosively reported last week. Business Insider also quotes a Microsoft employee who says the biggest disaster comes because, ‘All of a sudden Joe Blow can see the CEO's emails.’ Microsoft have denied that Joe Blow, Jim Zim or any of their colleagues have accessed any CEO’s inbox. The company has added that any data breaches are down to lax categorisation by administrators but the whole debacle contributes to a sense that the product is more Zune than iPod.

As Marc Benioff outlined in his response next year’s hype is going to be about ‘agentic AI’. We’ve all become familiar with the metaphor of treating AI as ‘an eager assistant’ but this will go a degree further next year when we’ll be able to ask products like Anthropic’s Claude or ChatGPT to take on regular tasks for us. You might ask an AI agent to pull a daily press round-up for you in your area of interest or to complete a marketplace analysis and turn it into a slide deck. Salesforce say that their AI agents will analyse customer chat and set about solving the customers’ problems without direction. In each case the work will be done by a semi-autonomous bot. Microsoft announced their own AI agents last week as part of a reboot of Copilot.

This episode of the Artificial Intelligence podcast is an essential listen to understand what this looks like and how it will impact our jobs:

This episode also references a full 5 hour interview with the founder of Anthropic which is also worth checking out (at 2x speed obvs). I certainly left with a sense that the people who have learned the talking point that ‘AI has peaked and it’s a bubble’ are talking out their harris, but true enough there’s probably been some overhyping of where we are right now.

---

“From his first day as chief executive, Nadella worked at the company’s Glengarry Glen Ross culture… in [his] first meeting with department leaders he wheeled in a cart loaded with copies of a book called Nonviolent Communication and gave one to each person.” The Wired piece on Microsoft’s cultural re-invention is a very good read, don’t skip this one.

A few years ago I did a podcast on the same topic:

• GenZ workers are continuing to record themselves being laid off (and if you hear the way it happens, you have to conclude they’re in the right posting these clips)

• Eat Sleep Work Repeat talked about coaching last week, this week the Association of British Mentors said that 7 out of 10 people felt that having a mentor had impacted their mental health

• I did a sponsored podcast about office supplier Fora, about their beautiful offices and the trends they’re seeing in the way that we’re working

• Chatbots that diagnosed illnesses outperformed doctors given the same information

• The world’s worst person is now bullying government workers to return to the office

• I love the gentle pace of the podcast Where are you going? so the discovery of adjacent work-related pod What’s it like to be? has been a delight

Outrage in the work chat

Edelman's Trust Barometer tells us that most employees see their company as the smallest big thing that they believe they have some impact over.

This can mean that, rendered helpless by global events or politics, workers look to their bosses to take a stand for them. How do we get this right? How can we navigate this world of outrage?

The last time President Trump was in power it led to employees becoming more active - who knows if the same will happen in 2025.

Karthik Ramanna talks us through the way to deal with outrage - and the actions that any leader can take to make the workplace a better place. His new book is out now.

Listen: Spotify / Apple / website