AI Reshapes Enterprise Software: Marketing Automation and Teamwork Tools Evolve

Never mind the copilots; make marketing the air-traffic controller! Consequential AI takes flight

If you have been looking for examples of how marketing can leverage the power of AI outside the world of writing content, Bloomreach is one you should look at. The omni-channel marketing automation platform provider announced the next generation of its platform at its recent Edge Summit, and Amanda Cole, Bloomreach CMO, filled me in on all the details, setting the scene thus: How do we remove the bottleneck of the marketer being able to create thousands or millions of individual journeys by really giving them a marketing team that's connected and working together and automated? To answer that, let's consider the work of a marketing team. You have writers, designers, workflow builders, data analysts, and more. Each one plays a role in building and delivering a marketing campaign. Cole explains that when Bloomreach was thinking about AI, it could have built agents to perform or assist with these individual roles. Customers are looking for results; they want to create, she argues, so what's the point of an AI agent that creates thousands of emails if it doesn't deploy them automatically? Marketers are at risk of dealing with a number of agents in their daily workflow, many of which won't talk to each other. And many that they will have to learn how to talk to each other to get the true benefits of AI and automation. Where is the efficiency in that? Bloomreach saw an alternative approach to using AI, one that can automate the end-to-end marketer workflow. Users prompt Bloomreach's Loomi AI with a request. One example from Cole: I want a Black Friday campaign for my high-value customers that gives them an offer they can't refuse. Loomi then builds the entire campaign. It creates the segment, writes the content and creates the design assets, builds the workflow and triggers, the channels, the variants, the reporting and dashboards, and all the analytics. Cole says the only thing it doesn't do is press the Send button (which it could do, but Bloomreach doesn't think we are there yet). Once the campaign is built, the marketer can view the entire workflow, and Loomi can take them on a tour that shows what it built and why. The marketer can then request changes at each stage of the workflow. Loomi will make the updates immediately so the marketer can review them. Things you can do - modify the segment to include another segment, change an image, add another channel, and the list goes on. What Bloomreach has built is what I have been waiting to see from marketing automation providers. Instead, most are creating agents that perform specific tasks within a platform. For example, HubSpot introduced several new agents, including the Content Agent, which helps create different content assets and improve existing content. Now, I'm not saying agents are bad. There's a huge opportunity here, especially when they can connect with other agents to create end-to-end processes. But what Cole described is exactly what I said marketers needed when I asked: Instead of relying on pre-built reports, why can't marketers ask questions and have AI do the hard work of analyzing and pulling everything together? And then enable the marketer to select which insights to act on and help create the framework of a campaign to make that happen. Sure, there's content creation in there, but it's only part of what happens, and it's not the first thing. But as Cole points out, it's more involved than building the AI capabilities. She says marketing automation providers will struggle because most of them don't have a way to understand intent, KPIs, or outcomes. Those things are not required data inputs in most platforms. For example, when you create a campaign in HubSpot, you don't have to add a description or tag to describe the campaign type. Bloomreach has an advantage here. It has all the revenue data within their platform (Bloomreach is also an e-commerce and Customer Data Platform), so it has been able to create an internal AI that could intimate the purpose of each campaign based on KPI results. The firm was able to go back in its history and build a campaign-based taxonomy. Cole says this taxonomy is necessary if you are trying to trigger AI or prompt AI to deliver a specific type of campaign. And since marketing automation platforms don't have the lexicon to understand a campaign request, Bloomreach thinks it will keep a lot of competition from entering the market. Cole argues: We have the advantage of using every campaign we've ever sent to understand what KPI it was targeting and which one best achieved that KPI so that we can use those journeys and workflow optimization to assist in the building of the campaigns. Bloomreach is also building benchmarking into Loomi. Say a marketer wanted to run a re-activation campaign. Loomi can suggest some campaign ideas and provide benchmark data on those campaigns from across Bloomreach's entire portfolio, all based on anonymous aggregated data. The marketer then has Loomi build the campaign. If the marketer prefers to build the campaign themselves, Loomi will provide a set of recommendations based on aggregate workflows. Loomi can even run a test and tell you how likely it will work. Benchmarking is where Bloomreach actually started, and then it was realized that Loomi could build the entire thing. Cole explains: Where we see most marketers kind of stopping right now is like they're interested in 'How does my campaign look compared to others?', and then getting recommendations on how to improve their campaign. I think it's because our imaginations haven't expanded enough beyond what our job is going to look like in the future, and so we're like, How do I do my job today better?'. What I am certainly challenging our team to do is to completely re-think your job, not because I don't want you to have one, but because there's this tremendous amount of opportunity to be doing something more human and more impactful. And I believe that genuinely. Cole notes that each company's lexicon to describe campaign types (reactivation, re-engagement, revenue-driving) and segment types (high-value, high earner, most likely to) might be different, and Loomi will need to learn it over time. It may take one or two prompts to learn a company's internal language, but Loomi learns incredibly fast. Bloomreach refers to its AI as pilots. Cole said co-pilots are important, but Bloomreach is pushing something it pitches as "Consequential AI." It's AI that moves the needle on metrics. She argues that's not just a co-pilot; it's doing your job differently. With consequential AI, the marketer becomes the air-traffic controller, she claims: If AI is the pilot, we actually believe you become the air-traffic controller. So, instead of being responsible for building a campaign, which you maybe very efficiently, could build 100 campaigns in a week or more using AI, we actually believe you should use AI to build 1000s, millions of campaigns. And then, if you think about the view of an air-traffic controller, you're watching all of these campaigns, you're absorbing the big metrics. So where is it green? Where is it red? Where are we stalled? Where are the challenges and giving you that opportunity for those uniquely human decisions? Cole suggests that with Loomi, marketers can automate as much of the operational work of running a campaign as possible and focus on the uniquely human elements of connecting with customers. There's also no limitation to testing. A marketer could run thousands of tests simultaneously, and what should bubble up to the top is the insights about their customers. When AI is introduced to a platform, one of the questions that comes to mind is, what will it cost the customer? Considering the amount of work that Loomi can do for marketers, you would expect the price tag to be high. Cole says Blooreach is still working with customers to make sure it's performing the way they need it, so they aren't actually at the pricing stage yet. She believes that when you pay a subscription for a SaaS product, there's an assumption that the brand continues to build new capabilities with the latest tech and that it's available. She calls this table stakes expectation, but some capabilities will go beyond table stakes. The other consideration is the cost of building this capability. Connecting to all the different systems and making API calls is much more expensive (Bloomreach partners with Google, Nvidia, and Open AI). How do you build that cost into the pricing? Bloomreach is trying to figure out where this new capability sits. They believe they have built something truly unique in the market, and once they validate it with customers, it will be priced accordingly. The key, Cole concludes, is that this new capability has to deliver significant lift, performance, and engagement. This is a month of conferences, and with those conferences come product announcements. Many of those product announcements are around AI as martech providers build even more AI and automation into their products What Bloomreach has announced is where I think these providers need to go. The opportunity for marketers to pass on more of the operational aspects of setting up and running a campaign or activity will empower them to spend more time understanding the customer, building relationships, and finding creative ways to build awareness, attract, engage, and build lasting customer relationships. But this will not be an easy shift for marketing teams. Marketing operations have become a primary focus for many marketers because technology is constantly evolving, allowing them to learn and do new things. However, this has kept them from doing more important work. It's also not a bad thing for marketers who specifically do marketing operations. It's an opportunity to grow and learn more technology that can help the company.

Sitecore introduces new AI capabilities with Stream, creating an intelligent DXP

It was a little over a year ago that then Sitecore Chief Product Officer, Dave O'Flanagan, shared Sitecore's perspective on generative AI and the Sitecore DXP, stating: When we think about generative AI and how we are implementing that in our solution, we're really thinking about how do we introduce better efficiencies in the workflows of our marketers or in the interactions of our customers so that things become easier or more relevant or more simplified? At the time, those efficiencies focused on content production, personalization, and search. O'Flanagan said AI should support a feature rather than be a feature itself, and they spent a lot of time deciding how to integrate generative AI into the platform. A year later, O'Flanagan is stepping up as CEO of the company, taking on the role in March, unable to turn down the opportunity to step up and have a broader impact on the company. So what's next on the journey for AI on the platform, delivering what he calls the fulfillment of Sitecore's vision for an "intelligent DXP"? Sitecore's mission is clear, both in terms of the product and technology it is building and on its customers. Along with continued investment in its products and a focus on composability, a crucial part of Sitecore's success going forward will be its innovation around AI, said O'Flanagan. That's why it has introduced Sitecore Stream, enterprise-ready AI built on Microsoft Azure OpenAI service. O'Flanagan argues that Stream will help orchestrate a seamless marketer experience across all Sitecore Solutions, including XM Cloud, Content Hub, and the Experience Platform: We felt the time was right to really rethink, you know, how marketers work with our platform. The value they can get from our platform. It's not just about bolting AI onto the platform. It's about kind of rethinking how marketers work and how we can help them be more productive in our tools and in our capabilities. Sitecore wants to provide marketers with "unrivaled parallel productivity and growth gains." What does that mean exactly? Well, to start, Sitecore announced three new AI capabilities. O'Flanagan said companies will ask themselves why they should buy generative AI capabilities from Sitecore when they could just use ChatGPT. The answer is simple: domain-specific language models. Sitecore believes in domain-specific language models, especially for enterprise customers. You take all the brand elements -- look and feel, tone of voice, copy, content, and imagery -- and train the model. It becomes the cornerstone of everything a marketer does across the Sitecore DXP. Each product in the Sitecore DXP is now enriched with new AI capabilities. For example, the Content Hub offers the following new AI capabilities: O'Flanagan explains that the AI technology and the content a co-pilot creates are designed to augment and assist marketers as they work through specific tasks. Sitecore co-innovated with its customer, Nestlé. Together, the two companies considered how Sitecore could help Nestlé and other customers) leverage AI to be more productive in marketing operations. Two areas that surfaced included co-pilots for brand and briefs. These two areas were important for Nestlé because they helped onboard new marketers more effectively. The Brief Co-pilot enables marketers to create briefs quickly, breaking them into specific tasks that are then shared with marketing and agents. Briefs are also shared with agencies, campaign teams, the web ops team, and others. Often, they aren't written correctly, and there's a lot of wasted time and effort, O'Flanagan says. The Brief Co-pilot reduces the time it takes to create a brief from days or weeks to minutes. It leverages the Brand Co-pilot to ensure the brief adheres to the company's brand and ensures everyone can work efficiently to deliver the campaign. There are also co-pilots for DAM, XM Cloud, CDP, and personalization. AI-assisted marketing workflow The third AI announcement focuses on what Sitecore saw as a gap in the marketing stack. Coordinating and collaborating within marketing teams and with external parties (like an agency) can be challenging at the best of times. How can AI help marketers better manage their work? O'Flanagan argues: If you think about a world where marketers are working in collaboration with agents, and we've got hybrid workflows, where humans and agents collaborate together to achieve goals. We've now created this layer in the Sitecore DXP that enables marketers to be able to do that. I don't think we're there in terms of AI being able to be - generative AI specifically - being able to fully automate some of these tasks. But with this new capability from Sitecore, you can confidently have a human in the loop as you look to address some of these tasks. Sitecore now includes workflows that automate repetitive tasks to help accelerate execution. Some examples include creating content, building a web page, or AB testing a call to action. Consider a marketer who needs to create a set of images and launch an AB test. AI can create the variants and push them through an approval process with marketers. Once approved, they can post the images to production. O'Flanagan said Sitecore wants to be able to safely apply all of these next-generation capabilities and technology in a way that is compatible with the enterprises it wants to work with. I asked O'Flanagan why co-pilots and not agents. He posits that co-pilots work alongside the marketer. They suggest things, or the marketer can ask for suggestions. An agent takes action on a marketer's behalf. The difference is augmentation versus automation, he says. As such, we are seeing a rapid evolution towards agentic workflows with humans in the loop, he suggests: I would say there were first-generation LLM capabilities that we all added, and all of our products have those capabilities integrated deeply now. So, if you want to translate some content, rewrite the content, generate some content based on a prompt, all of that is already natively integrated into the Sitecore portfolio. We see this as the next iteration, or the next evolution, where you take the AI, you wrap domain-specific capabilities, and then allow that to be provided to the marketer in an integrated workflow. Ultimately, the AI shouldn't be as dominant in the workflow. It's just going to be there assisting the marketer when and where they need it. I think the best examples of leveraging this type of advanced technology is where it's just seamlessly assisting and guiding the user along the journey. As CEO, O'Flanagan sees the bigger picture. He said technology is part of the solution, but it's also about Sitecore's services and support proposition. How do you set expectations in the sales process so that customers are comfortable they will get the ROI they need from a partner/vendor like Sitecore? O'Flanagan sees his role as being to help customers succeed. And the company is doing something right. It just passed US$500 million in annual recurring revenue, suggesting: We're at meaningful scale in the market. Along with the AI announcements, Sitecore also announced that most of the products in the Cloud DXP are HIPPA-ready (Health Insurance Portability & Accountability Act). These solutions include XM Cloud, Content Hub, Customer Data Platform, and Personalize solutions, he says: It's a build on all the work we've been doing from a compliance perspective internally, which means we can cater to heavily regulated industries. We can bring the best technology to these industries, which, sometimes, there's compromises to be made. If you have to deal with such heavily regulated industries, they might not get the newest tech, with the newest hosting, with the newest capabilities. I think we can solve that. According to Gartner, by the end of 2025, 30% of generative AI projects will be abandoned after a proof of concept. The reasons? Poor data quality, inadequate risk controls, escalating costs, or unclear business value. If a company doesn't have a clear roadmap and plan for implementing genAI technologies, this makes sense. Sitecore seems thoughtful about integrating generative AI capabilities into the platform. I appreciate that we didn't hear about a number of new standalone agents working with each product in the portfolio. Instead, we are seeing co-pilots who are brand-aware and help marketers perform their work more efficiently. This approach will be the right way to move forward for many because it allows marketers to slowly learn how to adapt AI capabilities in their processes. But at some point, and probably not too far in the future, marketers will need to look closer at automation and agents or co-pilots that do much of the work for them, allowing them to do the work only humans can do. And there's plenty of that to go around.

Celonis launches AgentC - 'Process Intelligence essential to any enterprise AI strategy'

Artificial Intelligence (AI) has been promising (threatening?) to become widely useful for citizens and businesses for a number of decades now. However, despite a few false starts, it wasn't until the launch of ChatGPT back in November 2022 did it really start to capture the imagination - and attention - of the masses. OpenAI's use of Large Language Models (LLMs) were, in part, successful in garnering widespread adoption because of ChatGPT's ability to give the impression of intelligence via conversation, by using publicly available data on the Internet. However, as we have seen time and time again, ChatGPT may be a useful tool to summarize some content, to bounce ideas around with, or to tidy up an essay - but the more specialized the need, the less effective it becomes. This is particularly true when it comes to business requirements. Enterprises are incredibly complex organisms, making use of multiple systems, various data identifiers, and wide arrays of siloed data. Technology has become increasingly useful in enabling the operation of business, but it's arguable that even the most sophisticated companies out there still rely heavily on the institutional knowledge of its employees. This is why over the past year and a half we have seen an onslaught of technology vendors attempt to solve the 'enterprise AI problem', with launch after launch of generative AI tools that make use of company-specific data or RAG technologies to make advancements in LLMs more useful. However, often the problem is that these tools are system-specific - a CRM platform may be able to help a customer service representative become slightly more productive by offering summarized call notes or recommended actions for next steps, but that doesn't quite equate to enterprise-wide organizational change. This is where Celonis believes it can play a significant role. Celonis started by offering process mining capabilities to organizations, focusing on individual processes (such as Accounts Payable). However, in recent years it shifted towards object-centric process mining, which took its process mining abilities from 'two dimensional to three dimensional'. In short, the launch of the Process Intelligence Graph allowed companies to map beyond single processes and create a 'digital twin' of their organization. The best way to think about it is that whilst process mining is very effective in following a single object through an organization (such as an 'order), it fails to take into account other objects that impact its progress (such as production, shipments, procurement etc). The outcome of this Process Intelligence Graph, and why it is relevant to the AI conversation, is that it provides a unifying data layer across any enterprise systems. In other words, it provides AI with the context of how your organization is operating - which, in theory, could be incredibly valuable when you're asking an LLM a question that requires a deep understanding of your business data. This context, provided by the Process Intelligence Graph, is what has enabled the launch of Celonis' AgentC today at its user event in Munich. AgentC is a suite of AI agent tools, integrations and partnerships that will allow users to develop AI agents in a number of leading platforms. Taking to the stage during his keynote at the Celosphere event, Celonis co-CEO and co-founder, Alex Rinke highlighted the ineffectiveness of a tool like ChatGPT in the enterprise - providing it with a question that may seem fine to a lay person, but one that didn't make much sense. Rinke said: Like every tech conference in 2024, we are going to kick things off with AI. Of course, the buzzword of the year. I want to show you something. We are going to bring up ChatGPT here on the screen and we're going to ask it a question that you might ask ChatGPT if you hadn't had your coffee in the morning. 'How do we handle excess stock levels and safety stock levels in HR?' [ChatGPT] is coming up with a very elaborate answer here, amazing buzzwords, like 'cross training' and 'skill development', stuff like that. But obviously all of you know that none of this makes any sense, right? The question doesn't even make sense. And ChatGPT is generating some rubbish here, because clearly it's missing something. It's missing very, very important business context. Rinke said that processes are the "beating heart" of a company and where organizations can find the most value and the fastest level of change. However, in order to do that, departments and their respective systems all need a tool that allows them to speak the same language - what Celonis sees as the Process Intelligence Graph. Rinke added: What we're missing is a shared understanding of how the business runs and how we can improve it. Efficiency is low, people are struggling and value is lost...but not anymore. Enter Process Intelligence. Process intelligence can be the connective tissue for your business. It gives you a real-time digital twin of your operations and also catches your unique business context. It's completely system agnostic, without any bias, and it provides that missing common language for understanding and improving how your company operates. And it's this context that Celonis sees as being critical to the advancement of AI in the enterprise: It connects you to your processes, your teams to each other, and emerging technologies like AI to your business. You're finally going to see AI that knows how your business operates, because clearly ChatGPT doesn't. As noted above, today Celonis has launched a suite of AI agent tools, integrations and partnerships, dubbed AgentC, that utilize the Process Intelligence Graph to provide enterprises with context-driven AI. First platform integrations include Microsoft Copilot Studio, IBM watsonx Orchestrate, Amazon Bedrock Agents, and open source developer environments like CrewAI. The first pre-built AI agents are available from ISV partners, such as Rollio and Hypatos. In addition, as part of AgentC, Celonis announced AI Assistant Builder, which allows customers to create AI assistants and copilots within the Celonis platform. During the keynote, Celonis' VP of Product Marketing, Divya Krishnan, provided a demo that showed how a user could select which data and knowledge they wanted an LLM to have access to, as well as allowing freedom over which LLM to use. Certain teams can have access to certain data and the LLM then has the relevant, deep process intelligence to provide answers. For instance, we saw during the demo that the LLM was able to provide specific answers to 'How do I improve excess stock levels?'. Referencing the Celonis CEO's earlier attempt at using ChatGPT, Krishnan said: Process Intelligence is essential to any enterprise AI strategy because it gives AI the context that it needs to understand your particular business. Remember Alex's question to ChatGPT? If AI doesn't understand that safety stock isn't a relevant HR KPI, then it can't really do much to help you improve how you run HR. And unfortunately it can't do much to help you improve safety stock levels, either. With our latest release, you can now configure these process copilots in a matter of minutes. You can give each team, from Accounts Payable to logistics, access to the relevant information that they need, in a natural language interface. What makes these process copilots so powerful is the foundation on which they are built on - the Process Intelligence Graph. Process copilots are now available in public preview. Celonis referenced a few companies that are already making use of AgentC: Speaking to why Celonis is somewhat uniquely positioned to take advantage of advancements in enterprise AI, Krishnan pointed to the ability to the ability of the Process Intelligence Graph to go enterprise-wide: AI is increasingly trying to take over a lot of the manual tasks and activities that comprise a huge amount of work for the enterprise - but for it to do that, it needs to have that context. It needs to understand: how do the processes actually run? What's supposed to happen at what stage? Who is actually responsible? How are these things calculated? Is this good? Is it bad? That's the class of data that is proprietary for every company, but it doesn't live in any other system. That is not data that's available in an ERP. It's not data that's available in a data warehouse. It's not data that's available in BI tools. And it's not even there in process documentation, because in many cases, that becomes outdated as soon as it's put down on paper. So there is no system or tool today that has all of this context around how the organization runs. The data that we have from the Graph is really what can play a crucial role in powering enterprise, AI. A compelling pitch from Celonis and one that we've seen coming for a number of months now - it's clear that its AI proposition is now fully formed and being made available to users. The results will now obviously be seen in the use cases, but I can confirm, having sat in a number of customer sessions today, that this is definitely of interest to buyers. A number of the conversations, driven by a wide variety of customers in the breakout sessions today, were focusing on how Process Intelligence can support their AI ambitions. Now, discussions are different to implementations and live results - but that will take a bit of time. Hopefully this time next year those use cases will be front and center, once AgentC has had some time to embed.

UiPath Forward - where does agentic AI go from here, and is RPA still relevant?

So I wrote an epic missive on the absurdity of AI agent overhype, hopped on a plane to Vegas, to hear UiPath make splashy news announcements about... AI agents. How did that go over with me? A bit better than expected - and I'll explain why. How did it go over for UiPath customers? That's the more important question. UiPath Forward day one keynote saw UiPath founder and CEO Daniel Dines declare that agentic automation is literally "act two" for UiPath itself. (My colleague Alyx MacQueen covered this off in UiPath's Second Act - RPA combined with agentic AI to enable real organizational change). A flurry of news items bolstered this proclamation, including: But hold up - does this mean that RPA - aka rules-based automation - is irrelevant, or legacy? With millions of UiPath RPA automations in play, customers certainly hope not. UiPath's answer is that task robots still have a very important role to play. That answer is not UiPath protecting its install base from a hard truth. Rather, it reflects the both the possibilities and the limitations of today's AI agents. In his candid keynote remarks, Dines acknowledged: The biggest lesson that we had throughout these years is that automation is really hard. It's very hard to deploy automation at scale. It's not only about the happy path and our process, it's all about the exceptions that appear. So you have to code back to go back and recode the exceptions. You always need to monitor the bots, the user interfaces. Other applications are fragile. You need to have people always people in the loop to be there. Enterprise workflows bring complexities: not every invoice, for example, is self-explanatory to a bot (e.g. it may require a discount not indicated in the data). Over the years, Dines told attendees, UiPath pressed on this issue: One of the key component that we build during [Act One] was our orchestrator, and this is the core technology that is capable of orchestrating between robots and humans - monitor them, manage them, deploy them at scale. Dines arrived at the burning question: where are the pros and cons of RPA, and what AI agents do better? What made RPA successful in this phase was the capability of really running thousands of automations in a reliable way. We made RPA reliable and scalable at the enterprise level, but we always knew we are limited in our power. The limitation was always the unstructured part of a process we couldn't get to. If the information is in natural language, hidden in long, complex documents where the decision cannot be expressed in rules, [RPA is] really limited. But the problem doesn't end here: the unstructured data is often the most important element. If we are frank, and if we look at most processes, you will see there is a combination between structured parts and unstructured parts. And actually most of the time they are really mixed, one to each other. Most of the time, I would say the unstructured part is dominant. So we had to invent discovery tools to make recommendations to our customers. What's the best way to discover the structured part, isolate it and put humans in the loop for the unstructured part? Thus "act two," and the role of AI agents. LLMs may or may not resemble human brain function, I'd argue not - but their inventors certainly derived inspiration from how humans perceive the world. No questions about this part: LLM technology is vastly superior at dealing with unstructured data than RPA bots. AI agents are not a new thing, a semantical point I am notoriously grouchy about, but generative AI agents do open up new possibilities, not just in autonomy and orchestration, but in their ability to parse a wide range of data and advance a workflow - perhaps making something resembling a human decision along the way. But as Dines points out, these generative AI agents are probabilistic technologies. RPA bots are deterministic, but AI agents are not. Proper AI design will be crucial here, as AI agents bring a different set of pros and cons entirely: It's non-deterministic. It's you cannot predict the answer of Gen AI. It's simply impossible to predict the answer. So while we all understand that it's extremely powerful, its own nature makes it extremely difficult to use it in the context of an enterprise workflow, because enterprise workflows needs to be reliable, deterministic. And our job right now it's actually to make it exactly as I said, reliable and deterministic and capable of using in an enterprise workflow, and we are going to spend the next few years in order to make it happen. Just how deterministic and accurate agents will become - and when - is a subject of fierce debate, and for good reason. In my case, for example, I believe the underlying gen AI technology has an inherent probabilistic limitation. But whether I am right about that isn't important at this moment, because there are plenty of scenarios for AI agents where 100 percent accuracy isn't required. Via UiPath's agentic approach, different companies, project teams and use cases can determine when human supervision will be necessary. This is why I was less agitated by the AI agent festival at UiPath Forward. UiPath has greater clarity than most vendors about why conventional automation still has a role to play. Even better, they have a vast catalog of enterprise automations that customers can use (or build) AI agents to trigger. During his keynote spot, Chandra Gnanasambandam, Senior Partner at McKinsey, really drove home why task robots still matter: Agents need tools, and those tools will be robots, because an agent needs to break down a complex workflow into tasks and sub-tasks, and at the sub-task level, the only way to get deterministic is through robots. So that's why you can't have agent without robots. You can't have agent without LLMs. You can't have agents without an API layer, right? And so we really think it's an evolution of the AI architecture, but we also think it's agents plus LLMs plus robots. During my informal talks with UiPath customers, I found a strong interest in AI agents, and a basic alignment with UiPath's direction. And yet - this comes as no surprise - there are plenty of questions on how this will work under the hood. The customers I spoke with were also trying to tackle core automation problems with still-manual processes. They are trying to figure out how the software vendors they work with can play nicely - and whether agents across vendors will play nicely also. Marc Benioff's day one cameo appearance was a reminder of how vendors are crossing over on these topics. We've also seen plenty of action with UiPath's SAP partnership: UiPath Joins Forces with SAP to Accelerate Enterprise Automation. This also ties into UiPath's Anthropic integration announcement. How much choice will UiPath customers have? During our AI deep dive interview, Mark Greene, SVP & General Manager of Product Management at UiPath, addressed this: We talked about agentic automation today. Just like robots essentially orchestrate across multiple systems, we'll orchestrate across multiple agentic platforms as well. Whenever we're designing a new solution, one first core principle we work for is it need to work across all ecosystems. It needs to work across any model. That's because, in some cases, our customer wants to bring a particular model that they've trained or have specialized for. They may have a preference for Gemini versus Claude or whatever. Great - bring your own. Bring your own model is a key component. In the midst of the agentic news flurry, where should UiPath customers go from here? Greene responded: 2. Convert your unstructured content into relational, structured data formats: You can train models today using our product - and that's a differentiator. Our document understanding is the leading, most accurate document understanding capability. We believe that's a superpower for our agents that are going to be working on more unstructured content. You can start turning that into an output schema that both a robot and an agent can use. 3. Take advantage of Autopilot for Everyone, which is in GA right now as well: "Start with some of the assisted/ supervised use cases of agents as well through that." Some readers may be skeptical that UiPath had a different tone on AI agents. If so, check out this day one takeaways slide: "Agents cannot be trusted" - how often do you see that in the AI keynote slide deck? As for guardrails: "The guardrails for agents are people and robots" - a good way to put it. The fusing of structured and unstructured data is clearly a top payoff of gen AI for the enterprise, but we're not there yet. UiPath talks in terms of "controlled agency," which is a useful framing. Still, even Dines acknowledged that AI agency means taking more humans out of loops. What the limits of that? It's a potent question. Are we empowering employees, or laying the groundwork for a mechanistic, job-scarce dystopia? Right now, technical limitations will prevent the latter, but that won't last forever. Therefore, we need to take strong positions, as individuals, employees, and enterprises, on what kind of AI future we envision. Perhaps we should do that before jumping on the AI agent bandwagon? During his keynote, Dines used the example of an AI agent that generates and proposes air travel itineraries. Once that agent is "trusted," maybe it will be empowered to book trips under $3,000. Not my trips, I can tell you that much, but that might not be the case for some lower-maintenance travelers - or some companies. But these use cases make for fun debates, and emphasize how fast things are moving. "The tech is changing quicker than we can keep up with," one customer told us - and this customer is intent on pushing ahead with AI agents. UiPath's challenge will be to bring all customers along, not just the early adopters. And we haven't even considered the differences between cross-vendor agent workflows in the cloud, versus automations that include on-premise deployments. These are important questions - good thing we have a sit down with Dines coming up next. Other enterprise vendors are advised to take a hard look at what UiPath did this week. The day one keynote was more customer-centered than most, with the CEO offering an atypically candid look at how we got here. In the process, the tech took a major step towards de-mystification. Then, on day two, instead of bland motivational speeches or futuristic roadmap-gasms, we got a meaty discussion of AI trends from third party analysts and academics. This was invaluable because it provided a gut check on UiPath's own agent messaging. The day two keynote ended with a preview of where UiPath in headed - then UiPath attendees could head into TechEd deep dives for more details. This is how you deliver keynotes that actually add to an event, rather than serve as "email catch up hour," and readers know how rarely I say something nice about event keynotes, which I tend to perceive as legacy artifacts that take away from essential networking/learning time. Another UiPath strong point: openly addressing security concerns and the new attack vectors AI agents present. How did UiPath do on my vendor challenge to present customers kicking @ss with current products as well? Better than most, but here, UiPath can improve the balance. Soon after the keynote, I went to an impressive automation session with UiPath customer Cologuard, which is doing vital work on colon cancer detection: This is the kind of story that deserves keynote attention - AI agents can share the airtime. After all, it will be a few more years before vendors completely turn their keynotes over to robots. I'm out of time and space with details yet to cover; stay tuned for my UiPath AI under-the-hood special. Meanwhile, Alyx and I have piled up a slew of vivid customer stories. As they say, watch this space.

Team '24 Europe - Atlassian channels AI in the context of enterprise teamwork

At the beginning of the year, I mapped out some of the challenges digital teamwork vendors are facing from the rise of generative AI. Here's my take on how Atlassian is faring, having attended both its US conference in April and its Team '24 Europe conference event earlier this month. Atlassian still faces some challenges as it evolves its platform and product offering to adapt to the rapidly changing enterprise teamwork market, but it has some uniques that play to its advantage, too. There's no doubt that, as Atlassian's CMO, Zeynep Inanoglu Ozdemir, told us at Team '24 Europe, "AI is actually changing the fabric of teamwork." But that doesn't mean we have to throw out everything we've learned about enterprise teamwork and start over. While AI was a big focus at the event, the vendor also emphasized what it calls its System of Work, which grounds its products and technology platform in the bigger picture of how enterprises should organize work to perform at their best. This provides the vision and roadmap in which Atlassian sees its customers making most effective use of its core tools -- work manager Jira, shared knowledge library Confluence, video messaging app Loom and the new AI agent platform, Rovo. Having an opinion on how its customers should use its products isn't a new thing at Atlassian. Drawing on its own experiences and those of its customers, it started creating playbooks to help teams figure out how to perform better more than a decade ago, and more recently it launched an internal program to support a fully distributed work model and collect and analyze data on the results. This very deliberate approach to making digital teamwork actually work for its customers is a key strength. Three core principles to the System of Work were front-of-mind at this month's event. They are: One point to call out about Atlassian's mission statement for its System of Work is that it's targeted specifically at tech-driven companies. This is a consequence of its history, of initially serving software engineers and then expanding out to other functions across the enterprise. This inevitably skews its customer base towards those where software technology is at the heart of what they do, and therefore the product offering is tuned to serve tech-centric organizations. Obviously, there will come a point with increasing digitalization where that describes every enterprise, and customers are already drawn from industries that are embracing digital technology, such as financial services and retail. But the focus is on enabling processes where people have to collaborate around collective knowledge to achieve a goal. It's less geared therefore towards more structured, touchless processes such as factory floor operations or supply chain execution. Anu Bharadwaj, President of Atlassian, tells me: We believe we can help customers that are technology-centric -- companies that want to do digital transformation, which includes companies from insurance and retail... where they think that technology is important to their business. If it's a purely physical business with no technology component, that's probably not our segment. But any company that believes that technology is actually impactful, is important to their business -- which really covers the vast majority of everything today -- we believe that those sorts of companies are the companies we can help. The new Jira is core to this endeavor, as the focal point for tracking the progress of work. This means it has to be a tool that everyone is going to be comfortable using, and shake off its historic reputation of being unfriendly for casual users. This is why a new look was unveiled at this month's event, putting the existing functionality into a more appealing 'skin'. David Meyer, Head of Product for Jira, says: The problem is the perception of Jira as a developer tool, and the kind of the skin that some of that functionality is wrapped in. A lot of the way we're thinking about this is not, 'Oh, we need to build 100 new features.' We need to make Jira look and feel different and help customers get to the features that we already have. The work may still happen elsewhere -- Atlassian's integration into third-party teamwork tools and other applications has always been strong and gets an extra boost from new search and agent integrations via Rovo -- but by tracking it in Jira, teams bring it into the context of the System of Work and the shared goals of other teams across the organization. In doing so, Jira meets a growing desire to bring more order and automation to how work is managed across the enterprise, Meyer believes. He says: The imperative for Jira is, I need to make it a place where every team feels comfortable tracking their work. Then the other thread is, there's an exogenous trend where I think business teams, non-technical teams, are becoming more sophisticated in how they track their work. The new ingredient across these building blocks is AI. While an essential ingredient, its role in Atlassian's view is to supplement or augment the work that existing team members do, rather than replacing them. Bharadwaj says: Our fundamental belief is that humans-plus-AI is what will make up the teams of the future. And for those teams that don't have these AI-powered teammates in them, those teams are going to get left behind because of the power of these AI-powered agents to make sure that the existing humans in the team are a lot more productive. The focus therefore is on processes that have humans at their heart, rather than wholly automated processes that are designed to eliminate the human element. In this context, the role of AI is free up humans to have more time and reflection to do the analysis, collaboration and creativity that they excel at. It does this in two principal ways: Based on its own experience, Atlassian presented several examples of just how much opportunity there is for AI to remove the overhead of finding and absorbing information within typical enterprise work routines. For example, when employees sit down to compose their annual reviews, there's usually some new guidance each year that they need to read through and digest -- but human nature is that most people just launch into the process without diligently reading through the documentation. So now Atlassian has built an agent that reads all the documentation and provides guidance on demand, as people are working their way through the form. This both saves time and improves the quality of the reviews. Another example is an agent that explains the company's internal strategy in very clear, simple language. The agent is called ELI, which stands for 'Explain it like I'm five,' and takes the character of a pre-school teacher. The point here is that employees may not have time to delve into all the detail of company strategy, but it's still important for them to make sense of it in specific contexts. As Mike Cannon-Brookes, CEO of Atlassian, explains: We have 12,000 people around the world. They don't need to understand all the bits and pieces of our strategy, but they need to be able to ask questions and understand why we do this or that. There's been a fair bit of commentary recently about the move from AI as co-pilot, in which an AI assistant provides help alongside a human as they work, to AI as agent, where the AI assistant goes off and autonomously completes tasks. Atlassian is firmly in the AI as agent camp, although elements of its Rovo assistant also fulfil a co-pilot role, for example when Rovo Chat sits in a side-panel on the user's screen. But the direction of travel is towards full-fledged agents, which are effectively treated as co-workers within a team who fulfil specific tasks. A big advantage of agents is that they can harness the existing automation and knowledge structures already built into platforms like Atlassian's, so that instead of users having to figure out what to ask the agent to do or how to build a prompt, an agent simply becomes a digital personification of a specific automation or chain of automations that already exist in the platform. There's still a lot going on under the surface -- Atlassian has been building up its team of AI talent and is doing the grunt work of testing new AI models as they emerge to ensure its platform keeps pace with the latest developments -- but customers aren't exposed to that heavy engineering element, unless they choose to be. Creating an agent is just a matter of using a low- or no-code tool to assemble the relevant data sources and automations. This results in a significant increase in the amount of automation that can teams can achieve. Bharadwaj comments: I have yet to meet a customer that said, 'Oh, my engineers don't have enough to do.' So what happens in those cases? Software is not really demand constrained. Software is supply constrained. [If AI] means more and more people can write software, more software gets created, more things get created... More people are now in the position to create, versus earlier when you had to know how to do technical coding in order to create something. So I think AI really sets people free and gives them superpowers that weren't available in the past, to be able to do things that are brand new. This isn't just a matter of reusing existing automations, of course. In a keynote panel during Team '24 Europe, a speaker reminded the audience of the importance of finding new ways to do things differently. According to Johannes Siebzehnrübl, VP/COO, Delivery Excellence, at Arvato Systems, an IT specialist and multi-cloud service provider with 3400 employees worldwide that is part of the Bertelsmann Group: We need to challenge the way we're working today. We shouldn't just use this AI to give an existing process artificial intelligence, without initially rethinking, is this the way we want to work tomorrow? We may have to critically challenge and question the way we've been working the past years. I think that's the very important aspect everybody needs to ask themselves. Challenge yourselves, challenge the way you're working, and continuously do so. This has to happen, not every year, but at least every two, three years, as technology evolves, we need to do that. There's new features coming one after another, and with each one of those we need to rethink whether we can solve the problems that we had yesterday at this moment differently with the technology that is available. Atlassian is moving fast to bring practical AI benefits to its customers, grounded in the work it's already done to build up a strong digital teamwork offering. But it's not the only vendor in this space, and the likes of Microsoft, Salesforce, Box, Asana and others are also pushing hard. One thing that was noticeable looking around the show floor at Team '24 Europe was the lack of big-name systems integrators, who are often important partners in delivering large enterprise accounts, even though Atlassian has a thriving ecosystem of specialist partners, some of whom are also capable of serving large global enterprises. Also the ongoing need to transition many larger customers off data center products onto the cloud-native product set remains a distraction, but it's a useful reminder that many customers are still solving yesterday's problems let alone today's AI challenges. However Atlassian has a highly defensible market presence and finds itself in the right place at the right time. Its roots in agile, continuous development also give it some distinctive qualities that other vendors often struggle with: This distinctive outlook may be its trump card in the competitive landscape of digital teamwork.