Opening keynote · 9:10 AM · Wednesday, May 13, 2026
All right, so our opening keynote and the most exciting session of the day is going to be Phil first giving his keynote on "It's Time to Get Serious About AI." So please welcome on stage Phil.
Terrific. Good morning. Good morning, Manhattan. How are you doing? Terrific. So what a time for our industry, right? It feels like every week there's a bombshell. Every three days there's a bombshell. Every two days we're getting erased by Claude. It just feels like it's never ending. I've never experienced anything like this since about 1998, '99, when everything was dot-com. Now it's like all AI.
I want to spend the next couple of days really making sense of what's going on in the industry. We've got a great mix of people at this conference who buy, sell, advise on this stuff. And I'd like to focus on the hard truths, but also the positives. Like, where can we go with this? How can we embrace this? And also, where can we be realistic? Because this isn't all upside. There's a lot of difficult things that we need to address.
So a little bit about us at HFS. I think we've been famous in our industry for Shaping the narrative. So we coined a phrase, along with a firm called Blue Prism back in 2012, called RPA. You might be familiar with that. That took us into the world of technologies surrounding automation, data, and AI. And my favorite invention was the Digital OneOffice 10 years ago now, which was all about bringing together the front and back offices to drive real value across customers and partners and suppliers.
The generative enterprise hit the world in 2023, just after the release of ChatGPT 3.5. And what's been awesome about these conferences at HFS is a lot of you have come back every six months. We really can test the mood every time we come back, and it's like one of hype, then it gets one of reality, then it becomes one of where the hell next type thing. So really looking forward to just getting the mood, the feelings of people in the room here as we go through the conversation.
Services-as-Software, nearly getting on for two years now since we actually coined that phrase. We ended up trademarking it, which actually wasn't that great because everyone suddenly said, can we use it? Is it okay? We're like, yeah, fine. And then our AI-First Deal Labs, which were launched late last year under the tutelage of my colleague Saurabh, which have been a phenomenal success for us at HFS. We're working with a whole bunch of enterprises and tech firms, service providers, helping everyone figure out how to take traditional legacy-type engagements and move them into more agentified AI-driven contracts and how to build vision and outcomes. So this is something we're really enjoying. It's really driving our business this year. We have a whole team who just work on the Deal Lab, and some of them are here today as well.
So let me get into the dialogue today. What's happening? Where are we going? How do we prepare? And then we're going to have a leadership panel to really talk about it. But, you know, the global economy is just taking one massive bet on AI right now. It's so much tied to growth, and everything around the AI infrastructure in particular is really driving the market. We're expecting $700 billion in hyperscaler capex this year. 30% of the S&P 500 is made up of the Mag 7, the Magnificent 7. We think spending on services and software will hit $1.5 trillion in the next three to four years. Anthropic is going to go IPO apparently this year. They're already talking about $1 trillion in anticipated IPO. I think it's going to be even bigger than that. And then 75% of enterprises are looking to replace services with AI solutions in the next two to three years. So the impetus is there. Some really big numbers.
And let's not forget when you've got probably the most successful AI firm at the moment, NVIDIA. Their market cap is now nearly as big as annual US healthcare spending, and it's bigger than Germany, Japan, UK, and India's GDP. So you think it's arrived when we're looking at numbers this size.
So let's think about three things. The fact that anyone who says we've seen this before is a very dangerous sentence. Every previous automation wave hit a specific category of work. You know, mechanical looms hit textile workers, ATMs hit tellers, spreadsheets hit bookkeepers. You know, each wave had a boundary, and AI just has none. So right now we have a technology that can simultaneously handle analysis, writing, coding, research, and judgment, and there's no adjacent category left to migrate into. So the breadth of disruption is unprecedented.
The speed of change is outpacing every institution built to absorb it. So the industrial revolution played out over 80 years. The PC rolled through the workforce in 20. The internet transformed commerce over 15. With this current cycle, we're asking workers to maybe retool three times in their career, whereas in the old days it was you maybe had to retool once to change to do something else. We're facing this gap where the technology is moving so fast. Can our institutions, can our policies respond fast enough?
The other thing we'll talk a bit more about is there's problems at the bottom of the career ladder or the junior ladder where every previous technology, they did eliminate the rungs, but they left the ladder intact. AI is very different. It eliminates the apprentice mechanism by which juniors become seniors. So what's going to happen if we just stop hiring juniors in 2026? Who are going to be the partners of 2031? Who's going to evolve into the positions when us lot move into retirement and other things in life? So breadth, speed, the removal of the ladder are the big three things impacting the industry.
It was interesting seeing the chart earlier. We've got a bunch of really fresh research across the Global 2000 organizations, where we've spoken with over 500 AI decision makers. And what's clear here is only 13% have actually reached some sort of enterprise maturity with AI. The other 87% are still experimenting. I'm showing you this because we're taking these maturity categories and we're going to show you what each of these are doing. So here's low maturity. They're just applying AI in defined tasks and processes within individual teams. And in high-maturity companies, the 13%, they're managing AI as a standardized enterprise capability with continuous optimization. And in between, you've got various moves towards that. But the reality is everybody is still in experimentation mode.
Today, 9% of companies say they've got AI agents running end-to-end, but they're planning 38% increase in spending on agentic AI in the next 12 months. In two years' time, they expect a seven-fold expansion to 67% of them running agents end-to-end within enterprises. So there's a real move happening where right now we're at phase one. People are going to move very, very quickly in the next couple of years.
As we mentioned, there's a real lack of strategy in place. Only 14% have a clear one. The rest are in development and moving towards that. So it would be good to hear more from you folks on your claims that you're more in the 14% and less in the bottom ladders here. Leadership is not aligned on where companies are betting. So 45%, only 45% actually share a definition of what agentic AI even is. 54% share a common view of AI's long-term role, and 77% say senior leadership doesn't even know how to get there. And this is coming from leaders driving these initiatives across our biggest 2000 organizations in the world.
On top of that, let's not ignore what's going on with workforce reductions. I've just listed a whole bunch here, but a lot of this is driven by tech firms. A lot of them may have overhired a lot during pandemic years, et cetera. But you can see here most of these companies are making fairly sizable layoffs already, like Meta, Oracle, Accenture in there, Lufthansa, companies like that. And they're citing AI as the reason for the layoffs. When we look at our research across the Global 2000, and we look at those two maturity categories I just presented, on the least mature organizations, 28% are planning active workforce cuts, and none of them are looking to add headcount. The most mature organizations with AI, only 2% are planning workforce cuts, and 37% expect growth in some areas. So the ones who have gone through the evaluation, they've figured out how to rewire their organization around this, are looking to grow. They're not looking to cut. And I think this is a critical area we need to focus on as an industry because this is how interesting and how worrying this can get.
So across Global 2000, 52% of leaders expect to reduce roles in the next two to three years. And this is really where we're seeing the risk. There's 4 million people in the near-term risk. This is of 90 million employees across Global 2000. So these are roles where employers are already cutting, and AI is a genuine driver. This is happening right now. It's not theoretical. So this number could be as big as 4 million. That alone is a big number when you look across the large-scale organizations. 23 million are having what we call a structural risk. So these are AI-exposed roles whose employers haven't acted yet, but the capability exists. The issue here is are these cuts going to happen? Are they going to evolve with time as companies mature? And then you've got 63 million employees who we think are safe, at least in the next 24 months.
So what is wrong? People are terrified of AI. Only 17% of employees feel safe. 72% fear being judged if their AI experiments fail, and don't have that psychological safety to try and make this work. 52% don't feel safe trying new things with AI. People are worried. Why should I put my job on the line to try this stuff out? I don't want to do anything different.
It takes me to something, I call it the AI velocity gap. We call it the individual Sunday experience. For ourselves personally, this stuff has had a massive impact on our personal lives. I remember a year ago I had to ask our audience who's got a subscription to ChatGPT. Now I'm pretty sure everybody here has, right? There's a zero-friction adoption going on. You can connect your Gmail, your calendar, your OpenTable. Yeah, I'll connect my Claude to my calendar and whatever. I don't care. I'll take the risk. I have tolerance for imperfection. There's immediate ROI. It just saves time on routine tasks. I can put that proposal through Claude and it does a nice restructure for me. I don't need a business case to use this stuff. I can experiment rapidly. I can test. I can iterate. I can adopt. I can make mistakes.
But then I go to work on Monday morning. Everything's siloed in SAP and ServiceNow. Getting unified access is a nightmare. So much is stuck in email chains. Knowledge is tribal. There's compliance paralysis everywhere. God forbid I can use my personal LLM account. I have to use the corporate one. And there's got to be big security protocols around how we operate with this. And then there's a big governance vacuum in companies. How do you audit this? Who is liable? Who's going to get fired if this all goes wrong? So your best employees are already AI augmented, but your enterprises, they're still forming committees. They're still making business justification out of this rather than actually going at it.
My concern is top talent aren't going to work for companies which are just so behind. You're just going to look at this company and think, companies with low AI maturity, high propensity to fire, high propensity to focus on cost. Who wants to work in those companies? I'm going to go work for someone that's embracing what's happening, is being honest about it. So we actually took some elements of the velocity gap and ran them against those low-maturity companies and the high-maturity ones. And you can see in terms of execution speed, the high-maturity companies can deploy in weeks, not months. Their decision advantage, they can 82% report faster decisions. In terms of customer impact, 88% of mature AI adopters are seeing double-digit CX gains. So the ability to tie AI to revenue is massive. 0% of low maturity are seeing any revenue impact versus 41% of high maturity. And in terms of ownership, this is another big one here, only 3% of mature organizations have unclear ownership of AI, where 25% of immature, they don't know who owns this stuff. It's just sitting around the company. No one's actually stepping up and really taking control.
I have a question. Do animals, like, do they want to know things? Why is blue, blue? Why is anything? Why is… Wait, the really old people, what did they wonder about? Can machines ask questions? Like, real ones? Are some questions just forever? Why can't I stop asking why? I have a question.
But hey, this has been going on for the 30 years we've been talking about when we had this thirst for knowledge and now it's absolutely coming alive within our organizations. And, you know, the human dimensions are being reshaped by AI in the workplace. You know, and in terms of cognition, you know, humans are directing. Our machines are thinking for us, we're editing. And in terms of connection, it's like every conversation has the third voice in the room. There's AI in every meeting, summarizing, suggesting. It changes the social contract. In terms of confidence, we used to get confidence from what we knew. Now we get it from what we can ask, right? In terms of performance, doing is automating, and deciding is the job. For our identity, expertise used to be our edge, but the model has matched our expertise now. So identity is shifting. And then in terms of trust, when the model decides, who do you fire? Who's to blame, right? Who's responsible for this stuff? The whole work experience is just being redefined.
So where are we going? You know, it's like in two years' time, asking how to use AI will be like, do you use electricity? I very much doubt we'll be calling this conference an AI summit or anything like that.
So we're really moving from what I see as tools where we are today, where humans approve every output, every code completion, every search. I think this year there's much more focus around the second phase we call teammates, where AI executes tasks inside guardrails. It books the meetings, it runs the analysis. The next phase is, I think, very exciting, which is this operating layer where AI becomes the default surface for every workflow. People review by exception, not by step. You know, this very warm phrase, but the human is leaving the loop and is now overseeing the loop. It's starting to make decisions and running by intent. AI is the execution layer. And then the phase four, I'm trying to avoid the phrase AGI, but this is where, you know, in terms of the workplace, AI reasons across domains at human expert level. So the organization now reorganizes around AI, not the other way around. We start off the day with trying to shoehorn elements of AI into the way we operate. We then eventually start organizing around the way AI can help us operate.
So it's going to really eat into the middle management and the middle layer of the organization. Today, 47% of the Global 2000 operate in traditional hierarchies and only 13% expect to operate in the same way in three years' time. I've spoken a lot with a lot of colleagues, peers, and other organizations about the shifts to the model. We work a lot in the services industry, which is probably one of the biggest people-centric industries around, we're being so affected by what's happening with impact of AI on just delivering for clients. And you can see the old pyramid model on the left. This was all about a junior execution layer, a middle management layer, which would basically convey orders up and down the chain, that sort of thing. And then the senior layer, really, leaders and architects. I see us moving to initially a transition model, which is where AI elevates everybody. So middle management stops being this relay layer and starts becoming more like player coaches. They're not status brokers anymore. They need to get much more hands-on. Leaders get better and faster at making decisions, and juniors come in more AI-augmented. They're elevated. No more grunt work for them.
And they're moving to what I've coined, something called a doer model. And this is sort of where I think the future is unsettled on how organizations need to organize. But ultimately, this is about, I think, juniors coming in and spending more time learning from mid-level player coaches. My friend Frank D'Souza suggested, I think it's a great suggestion, that juniors should spend two to four years incubating and being trained before they're ready to face the market. And then you can bill them out at the same rates as the mid-level experienced players. But AI is at the fulcrum of the whole thing. But less of this sort of pyramid to triangle, I think we need a model where everybody's doing and working together, and there's a much flatter hierarchy. So everybody's developing similar skill sets and benefiting from each other.
There's even some examples today in the services industry where providers are already making shifts towards different types of models. Accenture, they removed 22,000 mid-tier roles and actually while growing revenues and increasing their bench, Cognizant has done something similar in terms of really focusing on the entry-level layers. There's a company I know, Persistent Systems, I don't know if they're here, but they've only ever operated in a more sort of smaller scale growth model because they're smaller and they can take on smaller, higher level engagements in certain regards. So the model is evolving. But the one thing that's absolutely critical about this model is you need to focus on investing in young talent. My fear is you lose your culture and you lose your identity if you lose that cadetship model and you just focus on hiring experienced people and trying to rejig them, retrain them, you need to have that focus at the junior level, at the fresher level. Otherwise, mid-market firms are going to come along. They're going to eat your lunch.
Because ultimately, the principles of consulting and services, for example, they've always been the same. You know, it's about scaling talent, scaling intelligence, and building learnability as a capability. The principles are the same. What's different now are the skills we need and the technologies that we need to understand are changing from traditional platforms to many different types of tools and needs, and I'll go into that shortly.
So talking about services, you know, for 30 years, actually longer than 30 years, we were selling services by the hour, but that clock has stopped. You can see here, really look at BPO contracts and IT services contracts. Three quarters of them are going to be renegotiated in the near future. There's a big shift happening. How these contracts are renegotiated depends on the relationships these companies have, the desire to move away from traditional models. Some enterprises just want to cut cost. Others are going to think about how do I get more value. It's all shifting. Services are becoming software. This is where the economics break. You can see the tech vision that we've talked about from staff augmentation through the platform-led and Services-as-Software where we really get to where the market is shifting. And as we evolve, the shift moves more to a machine-led model supported by people than a people-led model supported by machines. And growth becomes less linear. It becomes less tied to adding people. It gets tied more to how can you increase revenues with the existing people base.
Now, when Jeff Bezos goes to Wall Street and says, I'm going to double Amazon in the next six years and I'm not going to increase my staff quotient, their stock price rocketed, right? As much as we hate to say it, that is what investors want. They want to see how do we scale and grow with an existing cost base, rather than keep adding costs to the model. And as we said, this is the shift. You can see the Global 2000s are looking much more at agentic services, and 74% would like to shift into a Services-as-Software model as they evolve. We talked about a $1.5 trillion market opening up. You can see here how the traditional global software and SaaS markets are going to get eroded into these Services-as-Software approaches. And the big question now is, you know, what companies are going to dominate?
But before we get there, I thought we could talk a little bit about what we've termed the Services-as-Software flywheel. And we look at LLMs, our large language models. They are the accelerator. They accelerate reasoning and code, for example, and knowledge. Agentic AI is really the orchestrator for the organization about decisions and compliance. Vibe coding, however we look at Vibe coding, it's really about producing more for the organization, working service agents directly with business intent, being able to redesign technology programming with business intent, not just software intent. And then finally, the FDE model, the Forward Deployed Engineers, which is a topic of many, many conversations right now.
And what I really like about this FDE model, it's like embedding a Formula One pit crew engineer directly inside the driver's cockpit. It's not just maintaining the car from the garage. It's constantly tuning performance in real time while the race is happening. You know, consultants optimize organizational theory, whereas FDEs operationalize reality inside imperfect systems. Bad data, legacy workflows, compliance constraints. It's basically three skills blended into one person. You know, it's a business analyst. It's a software engineer. You know, it's a relationship specialist. So we're moving to a model where being a consultant, being a value-added executive is really changing, a little bit like civil engineering or careers like that, away from the traditional model, and being able to oversee all these pieces of the organization and how they come together.
And then it's really impacting the industry. We see three swim lanes with Services-as-Software as the market takes off. You've got the enterprise tech spend and the enterprise service spend. So you've got your SaaS traditional companies like ServiceNow, SAP. They're all trying to come at it from a software-led servitization angle. Who's providing that system of record for clients? If you're in the organization today, do you need to go into SAP for some data? Or can you go into an agentic layer and say, I need this, I need that, and that layer then goes into your back office applications to get what you need. These SaaS companies are fighting for relevance within their organizations.
We then have services coded as software, which is embedding proprietary IP into services via modular platforms, and virtually every service provider now is building a story around how they're building out these agentic-level platforms to service their clients. Who are you going to turn to? Who's going to be your first port of call? And then you've got native companies. We have a few of them here as well, companies like Rhino, Lyzr, Writer, et cetera. They're delivering real-time outcomes through native platforms themselves. So this whole market is really figuring out where to go. It's unrecognizable from even nine months ago, I'm telling you.
And then in the last 10 days, you know, we've had the big two frontier model companies come out and announce their own entry into services. Anthropic announced $1.5 billion of financing from mainly private equity and a consortium to build out their own sort of FDE-type organization. They say we're going to go after the mid-market and partner with these SIs for the high-end firms. But these companies are under so much pressure to get their clients taking on their products. They need help. You saw the data from earlier. The market is moving at breakneck speed in the next two years, and only 10% odd are actually moving fast enough to actually invest in these platforms. And OpenAI then responded, I think on Monday, with their own $4 billion OpenAI deployment company. They bought a consultancy called Tomorrow and partnered up with a bunch of private equity and some service-for-service consultants. So McKinsey, interestingly there. And they're trying to out-big Anthropic with a bigger number. But the bottom line here is they're both going at this market. No one is safe from the exposure to the way things are shifting. And we need to figure out how to work with these companies.
Motivating other people's agents. The new alpha is companies that spend money on tokens. I'm putting you on a tip. A token improvement plan. — Okay, that is not a thing. — If we don't find a way to spend one trillion tokens… Okay, we can't spend one trillion tokens. We have failed. We're gonna run into quota limits. — Tokens, tokens, tokens, tokens, tokens, tokens, tokens, tokens, tokens, tokens, tokens, tokens, tokens.
So how do we prepare? I like this phase we're getting into is when your team stops saying, how do we use AI? And just start saying, this is how work runs. We finally crossed that AI Rubicon, as Julius Caesar did, where this is how we do work now. Let's stop debating how we did work. This is how we do work. We've got to stop optimizing for cost and really start thinking about optimizing for outcomes.
We're already suffering from rewarding employees for how many tokens they're burning through. So it was a great quote from a professor of economics, Charles Goodhart, which is, when a measure becomes a target, it ceases to be a good measure.
So to conclude, we talk about the Four Ps of AI value. And here we have a cross-section of 979 actual AI pilots and live engagements in play. This isn't a survey; this is actual AI in practice. And we judge them by productivity, prediction, personalization, and performance. So going into AI, when companies are running pilots, 54% are all about productivity. Speed, efficiency, scale, and cost. Once they get to production, we get to a state of what I call outcome ambition. So suddenly that cost obsession goes down to 19% and all the focus then shifts to how do you personalize and tailor experiences? How do you drive performance and strategic outcomes? And how do you drive prediction with AI? So what is clear here is once we've got past the how do we save some money doing this, it's, oh my God, how do we actually drive revenue and have intent-based leadership with this?
We've all heard of Jevons Paradox, which is the cheaper we can scale intelligence, the more we want. It's just natural to think, okay, I hired this company to do X, now they're doing that. They did a really good job. I can get them to do more. Or I can get my own people to now broaden out because they're becoming decision makers. They're becoming intent-driven leaders. They can start to shift their roles across one particular function. So we're all changing how we operate into a more intent-driven thought process. We'll talk a bit about tokenomics more, I think, in a further panel, but they're like an interim step, in my opinion. It's how do we get to some consumption-based metric and make it meaningful before we get to the real metrics.
Lead, just, you know, don't just oversee. These are six leadership traits I think are very crucial with AI, but what's interesting about this is these have been leadership traits for decades, but the ability to listen deeply, to demonstrate calm optimism, to navigate different styles of people, be a great communicator, make people accountable, make yourself accountable, amplify others and really recognize contributions and really seek their feedback. These are the skills that are going to make us successful in the next generation.
And then we get to, you know, how do we pay off these debts? We've got to stop obsessing around technology with this stuff, because you look here at what is holding back major enterprises from achieving their strategy this year. It's not really tech, people, or data. It's process. So the majority of companies today are really being held back by fragmented workflows, governance gaps, and unclear strategic direction. So how do we free these debts up? One of the things here is build unified governed data. Build a foundation for it. In terms of process, maybe just pick a workflow, name the owner, and redesign assuming agents. They handle 80% of it. Get your agents handling more of the process. In terms of your people, too many people hide their AI use. See them, protect them, invest in learnability. There's a fear of judgment we've talked about. Create a safe, psychologically safe environment for your people to flourish in an AI environment. And in terms of tech, modernize what AI has to talk to, APIs, identity, observability before agents. Less fragile tech today, less system failure tomorrow.
So that's where we're all shifting. So I'll leave you with one last thing, which Saurabh and co. will be talking a lot about over the next couple of days. Think about outcomes. Set up a Deal Lab. I won't go through all these pieces, but essentially, we've got to get away from location choice and cost savings and think about business cases built on those Four Ps. Think about data portability, adaptive ramp-ups. Think about variable budgets, model fees, compute storage upgrades. Think about data and model portability, AI model rebuilds. Start thinking about future ready and start thinking much less about the tired and expired that we're still slaves to in so many situations. Otherwise… Okay, thank you very much.
If you don't have an account, Register here |
With the exception of our Horizons reports, most of our research is available for free on our website. Sign up for a free account and start realizing the power of insights now.
Our premium subscription gives enterprise clients access to our complete library of proprietary research, direct access to our industry analysts, and other benefits.
Contact us at [email protected] for more information on premium access.
If you are looking for help getting in touch with someone from HFS, please click the chat button to the bottom right of your screen to start a conversation with a member of our team.
