How Atlassian Thinks About Developing New Uses For AI

This is the published version of Forbes’ CIO newsletter, which offers the latest news for chief innovation officers and other technology-focused leaders. Click here to get it delivered to your inbox every Thursday.

More proof that AI technology should be moving past the hype phase: Two AI pioneers were recognized this week with the Nobel Prize in Physics. In the 1980s, John Hopfield developed an artificial neural network, like what exists in the brain, that uses a method for saving and recognizing patterns, providing computers with the ability to essentially connect conceptual dots, writes Forbes senior contributor Tor Constantino. Geoffrey Hinton, often referred to as the “godfather of AI,” developed the 1980s “Boltzmann machine,” which could train neural networks to “learn.”

In the decades since Hopfield and Hinton first published their research, it’s become the bedrock upon which many technologists and programmers have built AI and machine learning platforms. But their selection for the Nobel Prize shocked many—and not just because their contributions have only a tenuous connection to physics. Hinton, who received the Turing Award in 2019 for his contributions to AI, actually left his job at Google last year so he could speak freely about the potential risks to humanity posed by AI systems. “It is hard to see how you can prevent the bad actors from using it for bad things,” Hinton told the New York Times after his resignation from the tech giant.

In an interview published hours after the prize was announced in the New York Times, Hinton said he hopes it will make people take his predictions about AI’s potential harm more seriously. But fears about the technology—which is largely unregulated in the United States right now—continue to thrive. After California Governor Gavin Newsom vetoed a bill last month that would’ve placed strict restrictions on expensive AI platforms, conversation about the right sort of regulation has picked up in many sectors, writes Forbes senior contributor Joe McKendrick. After next month’s election, AI regulation may become a topic for action in legislatures across the country and in Congress. But in the meantime, AI technology is sure to advance. After all, this year’s Nobel Prize for Chemistry went to David Baker, Demis Hassabis and John Jumper for their use of AI in designing and predicting the complex structures of proteins.

Thousands of companies worldwide use Atlassian’s collaboration platforms, which include Jira and Confluence, to unite teams and projects. As AI becomes more powerful, the enterprise software company is using its observations—both from its customers and its own experience—to bring the technology to its platforms. I talked to Atlassian President Anu Bharadwaj about Atlassian’s viewpoint and response to AI demand in enterprises. An excerpt from our conversation is later in this newsletter.

VALUATIONS

The roller coaster ride for Nvidia stock continues, with the company surging to become the world’s second most valuable, surpassing Microsoft, on Monday. At the close of markets on Wednesday, Nvidia was worth $3.253 trillion, as its days-long rally slowed slightly.

Nvidia’s stock and market capitalization has seen dramatic rises and falls in the last month and a half. After its most recent earnings report at the end of August—in which the AI processor company delivered record earnings and smashed expectations—its stock price tumbled because it didn’t project as much growth in the next quarter as some might have wanted to see. But since then, relatively small events have caused Nvidia’s stock to spike. At a Goldman Sachs conference in mid-September, CEO Jensen Huang said demand for Nvidia’s soon-to-come Blackwell processors was “so great,” leading the stock value to pop 8%. And another leap in value came last week when Huang said on a CNBC interview that demand for the new chip was “insane.” Monday started a new rally, courtesy of Super Micro Computer’s announcement that sales for its liquid cooling products alongside Nvidia’s GPUs were strong. And the stock climbed another 4% on Tuesday.

This seems to be proof that AI’s effect on markets is still strong. The selloff in August looked like investors were stepping back until they saw that Nvidia was worth the share price. But their behavior since then—buying up the stock whenever there is any potentially positive news—shows they are still overly eager about the technology’s possibilities.

TECHNOLOGY + INNOVATION

While Meta makes the lion’s share of its money from social networks, it’s no secret that CEO Mark Zuckerberg is angling to be one of the leaders in virtual reality and augmented reality. And while neither has taken off among the public at large just yet, Meta has continued to produce and improve tech tools to make them more accessible. Meta has worked with Ray-Ban on AR glasses since 2021, and officially launched the latest version, known as Orion, at its Meta Connect 2024 event last month. Meta’s Orion AR glasses have a built-in AI model that can generate information on demand, bringing 3D screen projections in front of a wearer’s eyes at any point in time, writes Forbes contributor Gerui Wang.

Days after the new Orion glasses launched, a pair of Harvard students went viral with a video showing how an older model of Meta’s Ray-Ban glasses can be engineered to identify random people on the street. The students were able to connect the glasses to a facial recognition search engine, then used AI to connect to online databases for home addresses and Social Security numbers. The students walked around the campus area, identifying people and pulling up information like their home addresses and parents’ names. The students intended the video as a demo to show what LLMs and data scraping can do today, as well as show people databases that may have information about them.

CYBERSECURITY

Passwords have been a necessary evil for computers for decades. They allow for user security, but they’re easily forgotten or stolen. More big tech companies have been moving away from passwords and toward passkeys, which are more secure on-device digital forms of authentication that require no memorization on users’ part, operating with secure logins to a larger service, QR code scans, or facial and fingerprint recognition. Last month, Google announced it was expanding its passkey program, allowing users to save passkeys to Google Password Manager from Windows, masOS, Linux and Android devices. Basically, for most devices on which Chrome might be used, a previously saved passkey can unlock a user’s account without requiring a separate password, writes Forbes senior contributor Davey Winder. Last week, Samsung made a similar announcement, Winder wrote in a separate article, making its 2025 models of its smart TV, refrigerator and appliances all work with passkeys instead of passwords.

BITS + BYTES

Atlassian’s Anu Bharadwaj Discusses How Internal Needs Inform Product Development

Thousands of companies across the globe use Atlassian’s collaboration tools—which include Jira, Confluence, Trello and Loom—to work together and manage projects. I talked to President Anu Bharadwaj about how Atlassian serves companies’ tech and AI needs, culture at the 22-year-old, founder-led company, and its growth prospects.

This interview has been edited for length, clarity and continuity. A portion of this interview was also excerpted in the Forbes CEO newsletter. The complete conversation is available here.

What are you seeing in the way of AI demand from enterprises, and what are you doing to meet it?

Bharadwaj: AI demand for enterprises has gone through a bit of a crest and a trough over the last year and a half. Atlassian builds a lot of collaboration software—mostly things like project management tools, knowledge management tools like company intranets, wiki or tools like Trello, Jira. We announced over the last year a number of AI-driven offerings in our portfolio. One, Atlassian Intelligence, really helps users in the workflow of project management, of writing content, writing articles or sharing collaborative documents with their teams. We brought a lot of AI features in the middle of that workflow to help summarize, edit your content, figure out how to draw information from multiple sources and answer a question that a new employee might have. We built that into the platform.

In addition, we have an AI-centric offering that we call Rovo, which really helps teams build agents which can automate and take the next step and action in a smart and intelligent way in any use case that they might be operating in.

I’ve worked at Atlassian for 10 years. When we launched it 18 months ago, it was the most positive reception that we’ve ever received across our products. We have millions of users on an active basis. Enterprises, especially, were battling with the problem of: We have a lot of information and a lot of data across the company, but how do we really make use of this? How do we make this accessible and actionable to all of our employees? The two offerings across Atlassian Intelligence and Rovo helped answer that particular question for our enterprises.

You said you’ve seen AI enthusiasm peak and also go down into a trough in the last 18 months. What has brought it down, and where are things right now?

Initially when the technology wave broke out, there was a lot of promise around what can AI do. It’s hard to predict timelines. But when people talked about AI, we talked about the full spectrum of what’s possible, so a lot of customers saw the extent to which AI can be deployed. We started talking about AI doing code generation. I heard a lot of questions about: Do you think software developers are no longer going to be required? Are we going to replace all the engineers in a company with AI? Which is honestly not a question anybody can answer, and we should be suspicious if somebody gives a certain answer.

But the timeline over which that happens, if it happens, is very long. When we started out with AI, there was a lot of hype around all of that appearing tomorrow. What is definitely possible is augmenting existing people that do certain jobs. There are really a lot of automated AI workflows that can help support people. A lot of enterprises thought about what’s the extreme possible application of AI? Now they’re seeing what are applications that are realistic and possible?

I would say what is possible and realistic today and in the short-term is still a phenomenal uplift. At Atlassian, we’re a 12,000 person company, and we’ve deployed our own virtual service agents powered by AI. Over half of the service requests that we needed to have our people answer are now handled by the AI agent. A customer who uses our virtual service agents has managed to reroute roughly 85% of the service requests they used to ask humans to handle to the agent first. So the productivity benefits are quite dramatic in specific use cases and specific arenas. I think enterprises are going through that calibration of saying, Okay, what is immediately actionable and where can I see evidence of this happening right now?

You are not only a company that creates these solutions, but you’re also an enterprise yourself. What kind of use of AI systems do you have internally at Atlassian? How did you develop them and how does that inform what you do as a company?

It’s one of our company priorities. We think about it in two lenses: What can we do to serve our customers and what can we do to use it ourselves as a public company? We’ve tried a few things. Some things have worked, some things have not worked.

Because we are a company that spends a lot on R&D, we have a 6,000-person-strong engineering workforce. We think about what does a software engineer really need help with on a day-to-day basis, and code gen has been particularly helpful for us in speeding up the coding part of what a software engineer does. But nearly 80% of [where] a software engineer or developer spends their time on is not coding. How do we help deploy AI to all of the other surrounding areas such that a software engineer’s life gets better?

We’ve built several internal agents, like Rovo, which is an agent framework. We’ve built agents that can help translate user experiences from designers directly into deployable code. We’ve built agents that can help prepare deployment environments that software developers can automatically deploy into CI/CD workflows. Because we are also a software development provider, it helps us in two ways: using it ourselves and also making it available for customers.

A second category is around customer support or service requests. This can take the form of external customers calling up service centers or filing tickets, or even internal help desk customers. We are now a 12,000-person-strong company. As new hires come online, they tend to submit requests like, I need a new laptop in this particular geo, or I have this question about payroll. A lot of HR, IT, those sorts of service workflows, that is an area where we’ve seen dramatic productivity benefits.

Then there’s this third category around collaboration and knowledge workers, which is largely around the blank page problem: How do you start creating content given a few cues, and then how do you iterate on top of it? I want to change the tone of it. It should sound as though Anu said this; it shouldn’t sound like it’s some kind of corporate comms coming in based on attempts.

A lot of that category of use cases around communication, editing, summarizing, drawing content from multiple sources—both content creation and knowledge discovery—has been the third category where AI has been particularly helpful for us. The thing that we’ve discovered [is] the number of SaaS tools, the number of products that enterprises use has been exploding over the last few years. Companies used to consolidate on an ERP platform, but then the era of SaaS began. A typical enterprise now uses 150 plus of these tools. Information gets fragmented everywhere.

FACTS + COMMENTS

Braintrust, a year-old startup that monitors AI products and helps them perform more accurately, recently closed a Series A funding round led by a16z partner Martin Casado.

$36 million: Size of the round, which brings its value up to about $150 million, according to Forbes calculations

80%+: Factual accuracy of an AI system that has used Braintrust for a few weeks, according to the company

‘A tool for helping everyone else to build AI software’: How founder and CEO Ankur Goyal described Braintrust

STRATEGIES + ADVICE

As economic indicators improve, companies might become more active in M&A. And while this impacts many areas of business, it’s important to remember that it also brings cybersecurity risks.

Executives constantly have decisions to make, and few of them are easy. Here are some ways successful billionaires have streamlined their decision-making processes.

VIDEO

QUIZ

In a motion to dismiss Elon Musk’s federal lawsuit against OpenAI filed this week, how did the startup behind ChatGPT describe the xAI founder’s case?

A. A money grab from the world’s richest person

B. The next chapter in a saga of regret for Musk’s early departure

C. Another opportunity for Musk to see his name in headlines

D. An increasingly blusterous campaign to harass OpenAI

See if you got the answer right here.