Transcript
00:00:00(upbeat music)
00:00:02Today, we're focused on closing the AI value gap
00:00:07and I'm thrilled to be joined by an expert on the subject
00:00:10and a Vercel customer, Dan Martinez,
00:00:13Managing Director at BCG Platonian.
00:00:16So Dan, welcome.
00:00:19- Thank you, Jane.
00:00:19Great to be here.
00:00:20- Awesome.
00:00:21Well, maybe to sort of set the table for us,
00:00:24BCG research found that only 5% of companies
00:00:27are generating substantial value from AI,
00:00:30while 60% are still struggling.
00:00:32What's creating this gap?
00:00:34Is it the technology problem, an execution problem,
00:00:37or something else entirely?
00:00:40- Well, Jane, if we look into the last three years, right?
00:00:44Since Gen AI basically started in 2023,
00:00:48lots of companies started with use cases and pilots, right?
00:00:52And I found that some companies were almost like competing
00:00:54for how many use cases they could get to.
00:00:57And sometimes they would get to 100 or 300.
00:00:59I've seen organizations have hundreds of these use cases.
00:01:03And I feel like ultimately
00:01:04people just got stretched through thin.
00:01:06Some of these ideas were very small.
00:01:08They were not what we consider process reimagination.
00:01:11They were not functional reimaginations of the organization.
00:01:15And then people just,
00:01:17I feel like they got lost in the shuffle.
00:01:18And some of these ideas,
00:01:19I think the business was aiming too low.
00:01:22Also, we found that some of these ideas were not,
00:01:27they did not involve capability build.
00:01:28So people were developing these use cases,
00:01:30but they weren't clear what is the change in job families?
00:01:33How does that change on upskilling?
00:01:35What is the impact on people?
00:01:36What is the impact on processes?
00:01:37So I feel like organizations were missing
00:01:39the bulk of the work, which is what we at BCG,
00:01:42we call the 10, 20, 70, which is 10% is stack,
00:01:4720% is data algorithms.
00:01:49And then 70% is really the bulk of the work.
00:01:53It's rethinking the businesses, rethinking tasks,
00:01:56how processes are different, who needs to be upscaled,
00:01:58how jobs will change.
00:02:00And I feel like 23, 24, it was just a lot of people
00:02:03experimenting, testing with these use cases,
00:02:07but not really thinking about,
00:02:09look, they have to go into production.
00:02:10They need to scale.
00:02:11We need to think about a whole bunch of stuff.
00:02:13So I feel like companies now are building the muscle,
00:02:18the discipline, the attention span,
00:02:20the leadership is looking at this.
00:02:22It's no longer, AI is no longer a technology project.
00:02:26AI is no longer a little experimental project.
00:02:30It's here to stay.
00:02:31It's existential risk.
00:02:33It's competitive advantage.
00:02:35- Yeah, that makes a lot of sense.
00:02:36I think your point on 70%, a lot of what I found
00:02:40in the work we've been doing in GTM is actually,
00:02:42a lot of that is even pre-production, if you will,
00:02:45of understanding what a best in class process
00:02:48ought to look like.
00:02:49And do you have all the content for that,
00:02:51having brought it through?
00:02:53So piggybacking on this,
00:02:54there's a phrase that's been coming up
00:02:55in the enterprise AI conversations,
00:02:57which is the shift from systems of record
00:02:59to systems of work.
00:03:01What does that mean in practice and why does it matter
00:03:03for how companies think about their technology investments?
00:03:07- Yeah, I first saw this concept in the article
00:03:09from VC in the Bay Area, where they talked about,
00:03:14you know, with the emergence of digital 20 years ago,
00:03:17companies moved from on-premise software to SaaS
00:03:20and moving to large enterprise packages,
00:03:22what we call systems of record, right?
00:03:24So if you think of Salesforce or ServiceNow or Workday,
00:03:28right, these are systems that hold a lot of corporate data.
00:03:31They have your customers, your orders, your deliveries,
00:03:36right, your financial data is in these systems.
00:03:39But then over time, we felt that people wanted
00:03:42to collaborate differently.
00:03:43And we've seen the emergence of more modern systems
00:03:46of engagement, for example, you know, Slack or Teams.
00:03:51Zoom, for example, and people are using these systems
00:03:53to engage, collaborate internally, collaborate externally.
00:03:56So it's almost like the user interface,
00:04:00to think from an enterprise architecture perspective,
00:04:02the UI has moved from the systems of record
00:04:04to the systems of engagement.
00:04:06And now what we're seeing with AIs
00:04:08is a new phenomenon altogether,
00:04:10which is the business logic of some of these systems
00:04:13of record are now moving to systems of work,
00:04:16and they're becoming agentic, right?
00:04:18So what we used to see as rules-based,
00:04:20deterministic types of features,
00:04:22now they're moving to probabilistic system prompts
00:04:25in these multi-agent systems.
00:04:28And of course, the hyperscalers are moving in that direction.
00:04:30They're creating lots of platforms.
00:04:32I mean, Vercel is in that scope as well,
00:04:35helping, allowing companies to very quickly,
00:04:37rapidly build these new agentic systems.
00:04:41And then we see companies like Salesforce,
00:04:42they're moving in that direction as well, right?
00:04:44They're building agent force as a capability
00:04:47and going to market with ready-made agents, right?
00:04:50And this is something that I feel like CIOs
00:04:52are starting to understand and grasp this new reality, right?
00:04:56Moving from these systems of record.
00:04:58How do I invest in these systems of record going forward?
00:05:01But then how do I build capability
00:05:03that allows me to shift these business rules
00:05:06into agentic systems?
00:05:07I feel like this is becoming more clear.
00:05:102025, 2026 is when we started to see
00:05:13organizations move to multi-agent systems,
00:05:16start to go from experimentation to production,
00:05:20building more resiliency, governance,
00:05:23all the architecture around it.
00:05:27And that's the pattern that we expect to see
00:05:29more and more in '26 and '27.
00:05:31- Yeah, I mean, I can bring that to life
00:05:33pretty specifically with Vercel,
00:05:35but how you describe it aligns exactly
00:05:38with what we've experienced here,
00:05:39which is we've got Salesforce, still a system of record.
00:05:44We started by building a singular agent
00:05:47to handle our inbound leads.
00:05:49So folks who fill out contact sales.
00:05:51In building that agent, we were able to go
00:05:54from 10 sales development reps down to one.
00:05:57That then formed the basis of a playbook platform
00:06:00where we now have multiple types
00:06:02of the sales development function.
00:06:04So event follow-up or hot PLG leads.
00:06:09That type of thing.
00:06:11So you've got all of those multiple agents running
00:06:13and then system of engagement.
00:06:16So a bunch of this stuff gets now piped into Slack
00:06:19or built custom workflow UIs
00:06:22'cause Salesforce front-end didn't necessarily represent
00:06:25those exactly how we wanted.
00:06:28So actually what you just teed up is precisely
00:06:30what we've seen play out in our first six months
00:06:34of bringing AI really deeply to go to market.
00:06:39- How do you help companies identify
00:06:42which workflows to prioritize?
00:06:44Vercel were very much working to avoid random acts of AI.
00:06:50So we found that the highest likelihood of success
00:06:53for agents comes from tasks that are a little bit more
00:06:57on the repetitive and deterministic side.
00:06:59So not a ton of cognitive load.
00:07:02The leading example I just got is a good one.
00:07:05Does that match what we're seeing?
00:07:07BCG, my knowledge is this language is like stop
00:07:10with the use case mindset and sort of open that
00:07:13and pilot purgatory I've heard a couple of times.
00:07:16So I think you're spiritually aligned
00:07:18with Vercel's random acts of AI.
00:07:20But again, how do you go from that rapid prototyping
00:07:23to picking the use cases
00:07:24that are actually gonna drive value?
00:07:26- Yeah, I think we're super aligned there.
00:07:28I mean, '23, '24, everybody was stuck in pilot purgatory.
00:07:32Learning, figuring out the technology, solving accuracy,
00:07:35hallucination problems, building rag applications,
00:07:40but ultimately realizing that it was very hard to scale.
00:07:44And I think people realized that it was hard to scale
00:07:47because the business,
00:07:48there's a lot of work on the business side, right?
00:07:50Retraining people, rethinking processes, et cetera.
00:07:53And I feel like we shifted from that use case pilot mentality
00:07:58to focus on value pools.
00:08:01And what are these big reshaped opportunities
00:08:05for organizations, right?
00:08:06So how does my servicing organization will be different?
00:08:10How will my finance function will be different?
00:08:13How will my supply chain function be different?
00:08:15So people start to amplify the scope,
00:08:18think process value chain level,
00:08:21picking specific examples in the value chain to drive,
00:08:25but really focus on a much bigger scope.
00:08:26And in a scope that's much more business-led,
00:08:29a scope that requires risk, compliance,
00:08:32legal to be involved to make sure that we understand
00:08:35all the ins and outs of this thing.
00:08:36And so we sort of moved away from use cases into value pools.
00:08:41Doesn't mean that companies are not using use cases.
00:08:43I still see that language happening,
00:08:45but we're moving to value pools.
00:08:46And we see, for example,
00:08:48some very clear value pools in the market.
00:08:50So for example, servicing, customer service, health task
00:08:52has been arguably number one area
00:08:55of where companies are using AI.
00:08:57We're starting to see a bigger emergence of startups
00:09:00in this space.
00:09:01Some are becoming well solidified in the market.
00:09:05AI for software engineering.
00:09:06I mean, this is a huge value pool for organizations.
00:09:09This is exactly where Vercel is squarely in
00:09:11as one of the leaders in the market,
00:09:13driving the charge here, driving the journey.
00:09:15I mean, I feel like we're just scratching the surface there.
00:09:18You know, the tools are gaining adoption.
00:09:21The engineering teams are building on top of it.
00:09:24I mean, some of these tools are becoming more integrated
00:09:27and embedded with the ecosystem and enterprises.
00:09:32This is actually one of the things I really like
00:09:34about Vercel, the fact that you guys already built
00:09:36lots of integrations that are very thoughtful about, right?
00:09:39So, you know, if companies need to do this
00:09:42on a hyperscaler, you know, they have to work
00:09:44through lots of hyperscaler services to pick from, et cetera.
00:09:47I feel like, again, we're just scratching the surface here.
00:09:49We're going to quickly move into using these technologies
00:09:52to build multi-agent systems,
00:09:54to build digital twins of organizations.
00:09:57And this is where we're starting to see
00:10:00the next future proofing of the organization, right?
00:10:02What's emerging at BCG is this ability
00:10:06to develop digital twins of processes,
00:10:10of functions, of the partners, right?
00:10:13This is such a scalable concept, right?
00:10:16If I'm, instead of focusing on use case,
00:10:18instead of focusing on value pools,
00:10:20can I create a digital twin of the organization
00:10:22and then simulate improvement ideas, right?
00:10:25And we're starting to like dip our feet in that
00:10:28in organizations where if an organization comes to us
00:10:32with a specific problem, we create this,
00:10:35it's almost like a reimagination AI
00:10:38that allows us to feed data into it
00:10:39and re-simulate tasks and processes
00:10:42and what if scenarios, right, at the enterprise level.
00:10:45It's a really interesting experiment.
00:10:47I mean, I feel like we're just now scratching the surface
00:10:49there as well, but hopefully that'll inform
00:10:52how we find these value pools in organizations, right?
00:10:56- This isn't exactly the point you were making,
00:10:58but on the thought of a digital twin,
00:11:01we have an internal data agent.
00:11:04You can think of it as like, take about a,
00:11:07like a data scientist analyst
00:11:09with about a decade of experience
00:11:11and it's sort of that level of capability.
00:11:13And this weekend, someone added that agent
00:11:16to the executive channel.
00:11:18And so we were all joking that this was, you know,
00:11:21the first agent promotion.
00:11:23But you know, we absolutely are doing that.
00:11:27We're pretty far along, I would say,
00:11:28on the data science side of things
00:11:31where you can actually see ways in which the agents
00:11:33that team is creating are in fact digital twins.
00:11:36You also started getting into sort of like, you know,
00:11:39how do you go from the prototype to production,
00:11:42touching on things like integrations,
00:11:44all the types of things that folks don't necessarily think
00:11:48about when you're prototyping, but you know,
00:11:50you don't want to have to go spin up 20 underlying services
00:11:53at AWS necessarily.
00:11:56So what are the best ways you've seen folks
00:12:00bridge that gap?
00:12:02- We're starting to bucket those gaps
00:12:04in specific archetypes for organizations.
00:12:06We came up with these four archetypes of AI agents.
00:12:09The first one is people are going to self-service
00:12:13the development of agents, right?
00:12:14And they're going to use,
00:12:16and maybe some people call them agents or not,
00:12:18but regardless custom GPTs or, you know,
00:12:23self-service tools where people are going to, you know,
00:12:26cloud skills, for example, and you know,
00:12:29people are going to use these tools to develop
00:12:32their own agents, connect with systems.
00:12:33Like for example, I have an agent that runs every morning,
00:12:37reads my email, sends me a summary of what do I need to do?
00:12:40What actions I need to take and sends me all the emails
00:12:43I need to respond, prioritize.
00:12:45Okay, I mean, that's a self-service agent.
00:12:47I run in one of the tools and it's helpful for me personally.
00:12:52But then we're going to see other types of agents
00:12:55that are built still by employees in organizations
00:12:58where they're built in tools like Microsoft Copilot,
00:13:02running in enterprise systems,
00:13:04running connected to tools like SharePoint,
00:13:08connected to data, et cetera.
00:13:09I mean, a bit more sophisticated,
00:13:11but still within the realm of employees developing them.
00:13:14Then companies are going to buy agents, right?
00:13:17And they're going to buy agents from Agent Force
00:13:19and whatnot, right?
00:13:20So we're starting to see, we're starting to do more,
00:13:22for example, market scan of agents, right?
00:13:25Just like we used to do for digital apps and SaaS companies.
00:13:28Now we're doing market scans for agents.
00:13:31And then the next one is where IT is going to come in
00:13:33and develop enterprise agents, right?
00:13:36And that's going to become much more science than art.
00:13:40It's going to become,
00:13:42there go to be a lot of rigor around these agents.
00:13:45We have to test them, develop them well,
00:13:48and there's going to be a lot more scrutiny
00:13:50around information security and policies,
00:13:55legal rigor.
00:13:57For example, responsible AI is going to be a big,
00:14:00important component there, guardrails.
00:14:02And then for these agents, we have an enterprise framework
00:14:06on how to develop these agents, right?
00:14:08This is where we see AI coding tools
00:14:10becoming a huge value for IT teams.
00:14:14I actually do think that when we think about buy versus build,
00:14:19solutions like Vercel and AI coding tools
00:14:21are going to enable IT teams to become very proficient
00:14:25at building.
00:14:26- Yeah, absolutely.
00:14:27I think we share a similar point of view on CIOs
00:14:30going from buyers of software to builders of software.
00:14:34I think a lot of the use cases we're seeing on Vercel
00:14:37are internal applications just as much as external.
00:14:40So if CIOs are now becoming software builders
00:14:42rather than just buyers,
00:14:44what does that shift from a role perspective?
00:14:47What's going to be new about the role of the CIO?
00:14:50- Yeah, that's interesting because on one side,
00:14:54this is totally elevating the buy versus build discussion
00:14:58and what does it mean for IT.
00:15:00We've seen companies like consumer companies
00:15:05start to hire agent developers, right?
00:15:07So these are no longer your typical machine learning engineer
00:15:11that may have a PhD in data science
00:15:15and knows Python really well.
00:15:18And I've seen a job description for one of these companies
00:15:22and it didn't even require Python, for example, right?
00:15:27So it's a new strange world that we're getting into, right?
00:15:30Now people are enabled and self-sufficient
00:15:33to develop their own agents.
00:15:35- Yeah, and so a lot of what you're describing here
00:15:37is really a central AI platform.
00:15:39And your research has shown that future built companies
00:15:43are 3X more likely to operate a central AI platform.
00:15:46Agents multiply across enterprise.
00:15:48What should that platform architecture actually look like?
00:15:52- We've been having lots of conversations
00:15:54with organizations around how to design this platform, right?
00:15:57And in the design, I'd say the design
00:16:01two years ago focused a lot
00:16:05on building simple rag applications, right?
00:16:08So it's all about make a choice on a vector database,
00:16:12make a choice on LLM that sits on your model garden,
00:16:16build guardrails at the application level,
00:16:18and you're good, right?
00:16:19And your biggest headache is accuracy problems.
00:16:23But that we've seen a departure from that thinking, right?
00:16:28And nowadays it's becoming much more complex, right?
00:16:31You need guardrails, not just at the agent level.
00:16:33We need guardrails at the orchestration level.
00:16:36You need to control not just for accuracy,
00:16:38you need to control for integration with core systems.
00:16:43There's a multi-layer way of thinking about security
00:16:46on these agents.
00:16:47So there's a lot to think about, right?
00:16:50CIOs are having to adapt their IT teams
00:16:55of skill, their architecture teams
00:16:56to be able to deal with this additional level of complexity.
00:16:59But that's what we need to think about
00:17:00when we go to multi-agent systems, right?
00:17:02Multi-agent systems is going to be a big step
00:17:04for organizations to be comfortable with,
00:17:06but that's what we see a big part of the value
00:17:09coming up in 26 and 27.
00:17:12- So you touched a little bit on the application layer there.
00:17:15If we're moving towards systems of work that we talked about,
00:17:18what role does the application layer play?
00:17:20Does the software that sits between AM models and users
00:17:23become more or less strategic?
00:17:25- I mean, for sure they certainly have a strategic role
00:17:29because they are the system of record.
00:17:31So they ultimately have the repository on data
00:17:36in the organization, right?
00:17:36So they will continue to be very valuable in that sense.
00:17:41They're also very valuable because they're going to provide
00:17:44those enterprise APIs for agents use in organizations.
00:17:49But the question is some of the business logic
00:17:53is moving from systems of record to systems of work.
00:17:58So it begs the question, what will happen to SaaS, right?
00:18:02We've seen some tech leaders saying that SaaS is dead.
00:18:06I'm not quite there yet, but I do think that they're going
00:18:09to become very strong databases
00:18:11with a very specific structure,
00:18:14with very specific control points,
00:18:17and they will continue to be valuable that way, right?
00:18:20Some of these companies are realizing
00:18:22that this trend is coming, they're moving towards AI,
00:18:25makes perfect sense, right?
00:18:27Some are more holding their fort and trusting
00:18:30that the wait-and-see mode a bit.
00:18:33But we'll see in next 12, 24 months,
00:18:37we're starting to see an emergence
00:18:38of these systems of work.
00:18:40Many of these are offering great opportunities for buy.
00:18:43I do think that SaaS companies will need to become AI first,
00:18:46right, instead of digital first.
00:18:49And that's going to take time,
00:18:50especially for some of the big ones.
00:18:52- So you mentioned there being a lot of opportunities,
00:18:54but you could also say that the AI vendor landscape
00:18:57is overwhelming right now.
00:18:58I think like most categories have 10 players in them,
00:19:01which seems like more than I'll probably be supported
00:19:04long term.
00:19:05What questions should enterprise buyers be asking
00:19:08to separate real capability from marketing?
00:19:11And how do they evaluate whether a tool
00:19:13actually deliver value versus become shelf-ware?
00:19:17- Well, for sure there's a technology fit, right?
00:19:20How will these companies, how will those agents run
00:19:24on a enterprise infrastructure?
00:19:27How are they integrated into that technology stack?
00:19:31How are they integrated with the systems of record,
00:19:34for example, right, that's an ongoing discussion.
00:19:37Then there's, then we ask questions around enterprise fit.
00:19:40For example, how do they manage compliance?
00:19:43How do they manage risk?
00:19:45How do they treat data privacy?
00:19:47Those are top of mind questions.
00:19:49You cannot, you know, it's a nonstarter
00:19:52at enterprise companies if they don't have a good answer
00:19:54for these types of questions.
00:19:56We look at cost, right?
00:19:59So that's the buy versus build cost, you know,
00:20:02and some of these solutions are very expensive, right?
00:20:06They charge the user per month level,
00:20:10and, you know, this is going to,
00:20:12in companies we need to allocate budget
00:20:14for these types of solutions.
00:20:16I mean, these solutions are coming.
00:20:17They're more expensive, but they're very valuable.
00:20:20And then we look at maturity of the company.
00:20:23As you said, some of these are new entrants.
00:20:25Many of these are still in series A, series B.
00:20:28Many of these have maybe 100 to 100 employees, right?
00:20:31So they're younger companies,
00:20:32and they're trying to get into an enterprise space.
00:20:34The enterprise space is very complex and requires
00:20:39a lot of attention, requires, you know,
00:20:41it's a long sales cycle.
00:20:43Some of these companies, they take six to nine months
00:20:46to onboard a new AI agent, right?
00:20:49That's pretty reasonable.
00:20:50I see that all the time.
00:20:51And companies are trying to figure out
00:20:54how do we fast track this process,
00:20:56but there's a lot of due diligence process
00:20:58to onboard one of these vendors, right?
00:21:00But I'm starting to see, interestingly,
00:21:03some of these started with small to mid-sized companies.
00:21:06Some of these vendors, these AI agents,
00:21:08they started with retail consumer.
00:21:12And I'm working with one of them,
00:21:14and this is gonna be the first quarter
00:21:15where the enterprise revenue tops the retail revenue.
00:21:19So we're starting to, again,
00:21:21we're starting to see the shift towards enterprise
00:21:24becoming the biggest customer for some of these solutions.
00:21:28- Yeah, seeing the same thing over here at Vercel.
00:21:31Well, Dan, thank you so much for joining us.
00:21:33This was a great conversation.
00:21:36For everyone watching,
00:21:37if you want to continue the discussion,
00:21:38please connect with Dan or me on LinkedIn.
00:21:41We'd love to hear what you're seeing
00:21:43in your own organizations.
00:21:45And if you're ready to move from prototype to production,
00:21:48check out the new V0 at v0.app.
00:21:51We just shipped some major updates
00:21:53that make it easier than ever
00:21:55to go from idea to deployed application.
00:21:58Thanks for joining our first shipped Q&A.
00:22:01We'll see you all next time.
00:22:03(gentle music)