Latest / Uphoff on Media Podcast / The Cost of Switching Is About to Go Away

Transcript

The Cost of Switching Is About to Go Away · Uphoff on Media Podcast

• 0:01
  SPEAKER_00: The cost of switching is about to go away.
• 0:04
  SPEAKER_00: Switching costs were the product.
• 0:06
  SPEAKER_00: Agentic AI just repriced it.
• 0:09
  SPEAKER_00: Uphoff on media by Tony Uphoff.
• 0:13
  There's a conversation happening right now in every Fortune 100
• 0:17
  boardroom,
• 0:17
  SPEAKER_00: every CIO suite and every IT budgeting planning session.
• 0:21
  SPEAKER_00: It goes something like this.
• 0:23
  SPEAKER_00: We know SAP is painful.
• 0:25
  SPEAKER_00: We know Oracle is expensive.
• 0:27
  SPEAKER_00: We know Salesforce has become a monster we barely recognize.
• 0:30
  SPEAKER_00: But what are we going to do?
• 0:32
  SPEAKER_00: You can't just rip it out.
• 0:33
  SPEAKER_00: The cost of switching is too high.
• 0:36
  SPEAKER_00: That sentence,
• 0:37
  SPEAKER_00: the cost of switching is too high,
• 0:40
  has been the moat protecting the enterprise software giants for
• 0:42
  over 25 years.
• 0:44
  SPEAKER_00: It's the reason SAP still runs global manufacturing,
• 0:47
  SPEAKER_00: Oracle runs global finance,
• 0:49
  SPEAKER_00: and Salesforce still runs global revenue operations,
• 0:52
  SPEAKER_00: despite decades of complaints,
• 0:54
  SPEAKER_00: workaround,
• 0:55
  SPEAKER_00: and usage concerns.
• 0:57
  Agentic AI is about to make that sentence a lot harder to say
• 1:00
  with a straight face.
• 1:02
  Legacy software vendors have confused the cost of switching
• 1:05
  with customer value.
• 1:06
  SPEAKER_00: The switching cost isn't just about the software,
• 1:08
  SPEAKER_00: it's about the knowledge,
• 1:10
  SPEAKER_00: the process,
• 1:11
  SPEAKER_00: the tribal memory baked into every customization,
• 1:14
  SPEAKER_00: every workflow,
• 1:15
  SPEAKER_00: every integration.
• 1:17
  SPEAKER_00: Agentic AI is about to transfer that knowledge somewhere else.
• 1:21
  SPEAKER_00: Three things Agentic AI will do to legacy enterprise platforms.
• 1:26
  SPEAKER_00: Let me lay out three scenarios.
• 1:28
  SPEAKER_00: They're not mutually exclusive.
• 1:29
  SPEAKER_00: In fact,
• 1:30
  SPEAKER_00: I expect all three will play out simultaneously,
• 1:33
  SPEAKER_00: but at different speeds in different organizations.
• 1:36
  SPEAKER_00: Together,
• 1:37
  they represent a tectonic shift in the economics of enterprise
• 1:40
  software that I
• 1:41
  SPEAKER_00: don't think the market has fully priced in yet.
• 1:45
  SPEAKER_00: Scenario one, the coordination layer makes legacy more valuable.
• 1:50
  SPEAKER_00: Here's the irony no one wants to talk about.
• 1:52
  Agenic AI will likely extend the useful life of SAP Oracle and
• 1:56
  their peers,
• 1:57
  SPEAKER_00: at least in the near term.
• 1:59
  SPEAKER_00: The real dysfunction in enterprise IT isn't that SAP doesn't work,
• 2:03
  it's that SAP doesn't work easily with the 12 other systems
• 2:07
  you're running
• 2:07
  SPEAKER_00: alongside it.
• 2:08
  SPEAKER_00: The data sits in silos,
• 2:10
  SPEAKER_00: the workflows don't connect,
• 2:11
  SPEAKER_00: the handoff between systems require human translators,
• 2:15
  SPEAKER_00: armies of them,
• 2:16
  SPEAKER_00: often called business analysts or systems integrators,
• 2:19
  SPEAKER_00: to move information from one platform to another.
• 2:22
  SPEAKER_00: Egenic AI is extraordinarily good at exactly this problem.
• 2:26
  SPEAKER_00: It can read the SAP instance,
• 2:28
  SPEAKER_00: query the Salesforce CRM,
• 2:30
  SPEAKER_00: pull the relevant ServiceNow ticket,
• 2:32
  reconcile the data and surface a unified view without requiring
• 2:36
  a multi-year
• 2:36
  SPEAKER_00: integration project or another implementation consultant.
• 2:40
  SPEAKER_00: I've called this the coordination layer in previous posts,
• 2:44
  SPEAKER_00: an AI native tier that sits above legacy platforms,
• 2:47
  SPEAKER_00: makes sense of them and orchestrates work across them.
• 2:51
  SPEAKER_00: The coordination layer doesn't replace legacy systems.
• 2:54
  SPEAKER_00: It makes them livable.
• 2:55
  SPEAKER_00: And for a lot of enterprises, livable buys another five to 10 years.
• 3:01
  This is why the smartest legacy platform vendors are leaning
• 3:04
  into the AI narrative
• 3:05
  SPEAKER_00: rather than fighting it.
• 3:07
  SPEAKER_00: SAP is building what it calls no-app ERP experiences,
• 3:11
  interfaces where a user simply expresses intent and an AI agent
• 3:15
  determines
• 3:16
  SPEAKER_00: which system to touch and what to do.
• 3:18
  SPEAKER_00: Oracle is embedding AI agents into its fusion suite.
• 3:22
  SPEAKER_00: Salesforce launched AgentForce.
• 3:25
  SPEAKER_00: They're not wrong to do this.
• 3:26
  SPEAKER_00: The coordination layer is real,
• 3:28
  and the platforms that integrate well with it will survive
• 3:31
  longer than platforms
• 3:32
  SPEAKER_00: that don't.
• 3:33
  SPEAKER_00: But,
• 3:34
  SPEAKER_00: and this is a critical point,
• 3:36
  the coordination layer creates a subtle trap for the legacy
• 3:38
  vendors because when
• 3:39
  SPEAKER_00: you abstract the user interface away from the underlying system,
• 3:44
  SPEAKER_00: you begin to extract the switching cost true.
• 3:49
  SPEAKER_00: Scenario two, agentic AI rebuilds legacy platforms from the inside.
• 3:54
  SPEAKER_00: The second scenario is longer term and more disruptive,
• 3:56
  SPEAKER_00: and it's one that legacy vendors are also betting on,
• 3:59
  though they may not fully appreciate the risk embedded in their
• 4:02
  own strategy.
• 4:03
  SPEAKER_00: AI assisted coding
• 4:04
  has already compressed software development timelines by
• 4:07
  meaningful factors.
• 4:08
  SPEAKER_00: Agentic coding,
• 4:09
  SPEAKER_00: where AI doesn't just auto-complete,
• 4:12
  SPEAKER_00: but actually plans,
• 4:13
  SPEAKER_00: architects,
• 4:13
  SPEAKER_00: writes,
• 4:14
  SPEAKER_00: tests,
• 4:14
  SPEAKER_00: and iterates,
• 4:15
  SPEAKER_00: compresses them by an order of magnitude more.
• 4:18
  SPEAKER_00: The implications for enterprise software are significant.
• 4:21
  Platforms that took decades to build and required armies of
• 4:24
  specialized developers
• 4:25
  to maintain can now be rebuilt iteratively and continuously at
• 4:29
  a fraction of the
• 4:30
  SPEAKER_00: cost.
• 4:31
  SPEAKER_00: The legacy vendors know this.
• 4:33
  SPEAKER_00: SAP is rebuilding its core ERP on a cloud-native architecture.
• 4:37
  SPEAKER_00: Salesforce is rebuilding its data model around AI-native principles.
• 4:41
  The question isn't whether they'll use agentic AI to improve
• 4:44
  their platforms.
• 4:45
  SPEAKER_00: They will.
• 4:46
  The question is whether they'll be willing to break the things
• 4:48
  that currently make
• 4:49
  SPEAKER_00: switching too painful.
• 4:51
  Every customization that took two years and four million to
• 4:54
  build is also a two
• 4:56
  SPEAKER_00: year,
• 4:56
  SPEAKER_00: $4 million reason not to leave.
• 4:58
  SPEAKER_00: When you can rebuild that customization in a week, the math changes.
• 5:03
  SPEAKER_00: The legacy vendors face a version of the innovators dilemma.
• 5:07
  SPEAKER_00: Their installed base rests on decades of accumulated complexity,
• 5:11
  SPEAKER_00: intricate customizations,
• 5:13
  SPEAKER_00: deep integrations,
• 5:14
  workflow configurations bake into the fabric of how their
• 5:17
  customers operate.
• 5:18
  That complexity has been their moat,
• 5:21
  expensive to replicate,
• 5:22
  painful to leave.
• 5:24
  SPEAKER_00: Agentic AI coding changes that calculus on both sides.
• 5:28
  It gives legacy vendors a powerful tool to modernize and
• 5:31
  rebuild their code bases
• 5:32
  SPEAKER_00: faster and more cheaply than ever before.
• 5:35
  But the same capability that accelerates their own rebuilding
• 5:39
  also lowers the cost
• 5:40
  of switching for customers and makes it easier for challengers
• 5:44
  to match what once
• 5:45
  SPEAKER_00: took decades to build.
• 5:47
  SPEAKER_00: If legacy vendors lead the charge on deploying egenic AI coding,
• 5:51
  they may be simultaneously modernizing their stack and
• 5:54
  dismantling the switching
• 5:55
  SPEAKER_00: cost advantage that protected them.
• 5:58
  SPEAKER_00: Some will thread the needle.
• 5:59
  SPEAKER_00: The ones who build genuine process power,
• 6:02
  who co-evolve with their customers in ways that create
• 6:05
  compounding workflow
• 6:06
  SPEAKER_00: advantage will survive.
• 6:07
  SPEAKER_00: The ones who are coasting on switching costs alone will not.
• 6:12
  SPEAKER_00: In scenario three, the Trojan horse.
• 6:15
  This third scenario is the one I find most interesting and the
• 6:17
  one that gets the
• 6:18
  SPEAKER_00: least amount of attention in analyst coverage.
• 6:21
  SPEAKER_00: Agentic AI doesn't need to replace legacy platforms to displace them.
• 6:26
  SPEAKER_00: It just needs to get inside them.
• 6:28
  SPEAKER_00: Here's how the Trojan horse could play out.
• 6:30
  SPEAKER_00: A company deploys an eugenic AI layer,
• 6:33
  SPEAKER_00: call it a coordination layer,
• 6:34
  SPEAKER_00: an AI co-pilot,
• 6:36
  SPEAKER_00: an automation platform that connects to SAP,
• 6:39
  SPEAKER_00: to Salesforce,
• 6:40
  SPEAKER_00: to Oracle.
• 6:41
  SPEAKER_00: Initially, it's additive.
• 6:42
  SPEAKER_00: It fills the gaps, improves the handoffs, surfaces insights.
• 6:46
  SPEAKER_00: The legacy system remain the system of record.
• 6:50
  SPEAKER_00: The AI is just the
• 6:51
  SPEAKER_00: intelligent interface.
• 6:53
  SPEAKER_00: But over time, the agentic layer learns the workflows.
• 6:55
  SPEAKER_00: It understands the processes.
• 6:57
  SPEAKER_00: It knows which data lives where,
• 6:59
  SPEAKER_00: which approvals are required,
• 7:01
  SPEAKER_00: which exceptions get escalated to which humans.
• 7:03
  SPEAKER_00: The institutional knowledge,
• 7:05
  SPEAKER_00: the tribal memory that made switching so expensive,
• 7:08
  SPEAKER_00: began to migrate from the legacy system into the AI layer.
• 7:13
  SPEAKER_00: The day comes when a CIO asks,
• 7:15
  SPEAKER_00: if all our workflow intelligence lives in the AI layer,
• 7:18
  SPEAKER_00: what exactly is our legacy platform doing for us that a lighter,
• 7:22
  SPEAKER_00: faster,
• 7:22
  SPEAKER_00: cheaper system of record couldn't do just as well?
• 7:26
  SPEAKER_00: This is the moment the switching cost calculation changes fundamentally.
• 7:29
  The new system doesn't need to replicate 20 years of
• 7:32
  accumulated customization
• 7:33
  SPEAKER_00: because the AI layer already has it.
• 7:36
  The migration isn't a rip and replace,
• 7:38
  it's a swap of the underlying data storage.
• 7:42
  SPEAKER_00: What took 18 months and 30 million now takes weeks.
• 7:45
  SPEAKER_00: I don't think this scenario is hypothetical.
• 7:48
  SPEAKER_00: I think it's already beginning.
• 7:49
  SPEAKER_00: The AI native procurement platforms,
• 7:51
  SPEAKER_00: the AI native finance tools,
• 7:53
  SPEAKER_00: the AI native HR systems,
• 7:56
  SPEAKER_00: they're not just building better interfaces,
• 7:58
  SPEAKER_00: they're building Trojan horses.
• 8:00
  They're positioning themselves to capture the workflow
• 8:02
  intelligence first and worry
• 8:04
  SPEAKER_00: about the system of record second.
• 8:08
  I was in the room in the early 2000s when the publishing
• 8:10
  industry first encountered Google.
• 8:12
  The search engine was framed as a discovery tool,
• 8:15
  a way to help readers find content.
• 8:17
  SPEAKER_00: Publishers welcomed it, even celebrated.
• 8:20
  SPEAKER_00: Traffic was traffic.
• 8:21
  What nobody fully appreciated at the time was that Google
• 8:24
  wasn't just indexing content.
• 8:26
  SPEAKER_00: It was inserting itself between the reader and the publisher,
• 8:30
  learning what readers wanted and gradually making itself the
• 8:33
  indispensable
• 8:34
  SPEAKER_00: intermediary.
• 8:35
  SPEAKER_00: By the time publishers understood what had happened,
• 8:37
  SPEAKER_00: the relationship that had always been theirs,
• 8:39
  SPEAKER_00: direct,
• 8:40
  SPEAKER_00: proprietary,
• 8:41
  SPEAKER_00: defensible,
• 8:41
  SPEAKER_00: belonged to Google.
• 8:43
  SPEAKER_00: The revenue followed the relationship.
• 8:45
  The agentic AI layer is doing the same thing to legacy
• 8:49
  enterprise platforms.
• 8:50
  SPEAKER_00: It enters as a helpful intermediary.
• 8:53
  It improves the experience,
• 8:54
  it fills the gaps,
• 8:55
  and gradually it captures the relationship.
• 8:59
  SPEAKER_00: The workflow intelligence,
• 9:00
  SPEAKER_00: the institutional knowledge,
• 9:01
  SPEAKER_00: the user behavior,
• 9:02
  SPEAKER_00: and the legacy vendor assumed was permanently theirs.
• 9:06
  SPEAKER_00: When that transition is complete,
• 9:07
  SPEAKER_00: the legacy system is what the publisher content was to Google.
• 9:11
  SPEAKER_00: Still necessary,
• 9:12
  SPEAKER_00: still providing the underlying data,
• 9:14
  SPEAKER_00: but no longer in control of the relationship that creates value.
• 9:18
  Publishers laughed at the idea that a search engine could
• 9:20
  disintermediate their
• 9:23
  SPEAKER_00: reader relationships.
• 9:24
  Enterprise software vendors are making the same mistake about
• 9:27
  agendic AI layers
• 9:28
  SPEAKER_00: sitting on top of their platforms.
• 9:30
  SPEAKER_00: The intermediary always wins.
• 9:36
  what the 16 ounce analyst gets right and misses,
• 9:39
  the smart folks at 16 ounce.
• 9:41
  recently published a piece arguing that AI will be the best
• 9:45
  thing that ever
• 9:45
  SPEAKER_00: happened to enterprise software.
• 9:46
  SPEAKER_00: Their argument rests on the durability of the classic business moat,
• 9:50
  SPEAKER_00: switching costs,
• 9:51
  SPEAKER_00: network effects,
• 9:52
  SPEAKER_00: brand cornered resources,
• 9:53
  SPEAKER_00: processing power.
• 9:55
  SPEAKER_00: They're right about most of it.
• 9:56
  SPEAKER_00: Network effects are real.
• 9:57
  SPEAKER_00: Brand still matters.
• 9:58
  SPEAKER_00: No one ever got fired for buying SAP.
• 10:01
  Process power as they define it is genuinely powerful when
• 10:05
  earned rather than inherited.
• 10:07
  But I think the piece underestimates how quickly switching
• 10:10
  costs are about to compress.
• 10:12
  SPEAKER_00: To their credit,
• 10:13
  SPEAKER_00: the authors acknowledge this directly,
• 10:15
  noting that AI is changing the friction and cost-benefit
• 10:18
  analysis associated with
• 10:20
  SPEAKER_00: switching vendors.
• 10:21
  But then the analysis moves on quickly,
• 10:23
  framing it as a net positive for the industry.
• 10:25
  Healthy competition,
• 10:26
  better products,
• 10:27
  faster innovation,
• 10:29
  true enough at the macro level.
• 10:31
  SPEAKER_00: What the investor lens misses is the operator experience.
• 10:34
  I've run companies that operated on Oracle,
• 10:36
  on Salesforce,
• 10:37
  and other legacy platforms.
• 10:39
  SPEAKER_00: I know what it actually costs to switch,
• 10:41
  SPEAKER_00: not just in licensing fees and integration hours,
• 10:44
  SPEAKER_00: but in the human capital required to manage the transition,
• 10:47
  SPEAKER_00: the productivity loss during the migration,
• 10:50
  the political capital spent convincing an organization to
• 10:53
  absorb the disruption.
• 10:55
  SPEAKER_00: Those costs are real.
• 10:56
  They are also increasingly solvable by the same agentic AI
• 11:00
  that's being deployed to
• 11:01
  SPEAKER_00: improve the platforms in the first place.
• 11:05
  SPEAKER_00: The switching cost was never primarily a software problem.
• 11:08
  SPEAKER_00: It was a knowledge and coordination problem.
• 11:10
  SPEAKER_00: Agentic AI is a knowledge and coordination machine.
• 11:15
  SPEAKER_00: All right, so what does this mean for B2B technology leaders?
• 11:18
  SPEAKER_00: If you're a CIO or CTO sitting on top of a legacy enterprise stack,
• 11:23
  SPEAKER_00: here's what I'd be thinking about right now.
• 11:25
  SPEAKER_00: Number one, the coordination layer is your immediate opportunity.
• 11:28
  SPEAKER_00: Before you make any platform decisions,
• 11:30
  invest in understanding what an agentic AI layer can do to
• 11:34
  improve the ROI of what
• 11:35
  SPEAKER_00: you already have.
• 11:36
  SPEAKER_00: The platforms that integrate well with AI agents will extend their value.
• 11:40
  SPEAKER_00: The platforms that don't will accelerate their own obsolescence.
• 11:44
  Number two,
• 11:44
  be honest about what your switching costs are actually made of.
• 11:48
  How much of the pain of leaving SAP is genuine platform
• 11:51
  capability and how much is
• 11:52
  SPEAKER_00: accumulated complexity that no one has bothered to modernize in 15 years?
• 11:58
  SPEAKER_00: Agenic coding can rebuild that complexity faster than you think.
• 12:01
  The number you've been carrying in your head is the cost of
• 12:04
  switching is probably too high.
• 12:06
  SPEAKER_00: And number three, watch Trojan horses carefully.
• 12:09
  The AI native vendors coming into your market aren't just
• 12:12
  offering better software.
• 12:13
  They're offering to absorb your institutional knowledge and
• 12:17
  make it portable.
• 12:18
  That's a different value proposition than any previous
• 12:20
  generation of enterprise
• 12:22
  SPEAKER_00: software challenger has been able to make.
• 12:24
  SPEAKER_00: Take it seriously.
• 12:25
  SPEAKER_00: And if you're building in this space,
• 12:27
  SPEAKER_00: if you're the challenger,
• 12:28
  SPEAKER_00: the AI native upstart,
• 12:30
  SPEAKER_00: the next generation platform,
• 12:31
  SPEAKER_00: understand the game has changed.
• 12:33
  SPEAKER_00: The incumbent's moat is draining faster than most people realize.
• 12:37
  The question is whether you can build genuine process power
• 12:40
  before the incumbent
• 12:42
  SPEAKER_00: replatforms around AI and builds it back.
• 12:45
  SPEAKER_00: The world will still run an SAP for the next few years,
• 12:49
  SPEAKER_00: but the reason it couldn't be replaced is quietly disappearing.
• 12:53
  SPEAKER_00: That's not a threat to enterprise software.
• 12:55
  SPEAKER_00: It's the biggest platform shift in a generation.
• 12:59
  SPEAKER_00: Okay, one more thing.
• 13:00
  SPEAKER_00: There's a version of this story where the legacy platforms win,
• 13:04
  SPEAKER_00: where SAP,
• 13:04
  SPEAKER_00: Oracle,
• 13:05
  SPEAKER_00: and Salesforce successfully embed agentic AI into their cores,
• 13:09
  SPEAKER_00: rebuild their platforms to the AI era,
• 13:11
  SPEAKER_00: and emerge even stronger moats than they had before.
• 13:15
  once their accumulated data advantage and process power
• 13:18
  compound with AI capability.
• 13:21
  SPEAKER_00: That's possible,
• 13:22
  SPEAKER_00: maybe even likely for the platforms that move fast enough,
• 13:25
  SPEAKER_00: but it requires them to do something that large,
• 13:28
  SPEAKER_00: mature,
• 13:28
  SPEAKER_00: PE-backed,
• 13:29
  and publicly traded enterprise software companies are not
• 13:32
  historically good at,
• 13:34
  SPEAKER_00: cannibalizing their own switching costs before someone else does.
• 13:37
  SPEAKER_00: I spent 35 years watching incumbents try to navigate platform shifts.
• 13:42
  The ones who succeed are the ones who accept that the moat they
• 13:45
  build is temporary
• 13:46
  SPEAKER_00: and start building the next one before they have to.
• 13:48
  The ones who fail are the ones who defend the moat right up
• 13:51
  until someone walks through it.
• 13:53
  SPEAKER_00: Agentic AI is already at the game.
• 13:57
  SPEAKER_00: Thanks very much for listening.
• 13:59
  SPEAKER_00: Please take the time to subscribe to A Puff on Media on Substack.