Artificial intelligence is no longer just a boardroom talking point for enterprise software buyers. In The Information’s latest Deep Research analysis, the pressure on traditional software is beginning to move from market fear into concrete procurement decisions, with companies already using AI to reduce hiring needs, bypass new license purchases, and replace parts of expensive workflows. The report’s central message is not that enterprise software is collapsing overnight, but that the old seat-based model is under visible strain as AI becomes more capable of absorbing routine work. That shift is already producing winners, losers, and a lot of uncomfortable questions for vendors that built their businesses on human-scale usage.
For most of the past decade, enterprise software enjoyed a remarkably stable playbook. Vendors sold seats, expanded through adjacent modules, and relied on the fact that growing companies tended to add more people, more processes, and more software licenses at the same time. AI disrupts that logic by attacking the one assumption the model depends on most: that every new workflow needs another human user, or at least another paid account.
The Information’s analysis frames this change as an early but meaningful pivot. Rather than treating AI only as a pilot-project novelty, companies are beginning to use it to rethink existing software spend, especially where repetitive knowledge work can be automated or compressed. That matters because software budgets are rarely cut all at once; they are usually reallocated one workflow at a time, until the cumulative effect becomes impossible to ignore. The report says nine recent examples show companies deploying AI to cut costs, though the impacts are uneven and still unfolding. al context is important. Earlier waves of automation mostly replaced manual labor in factories or standardized back-office tasks through rules-based systems and robotic process automation. Generative AI is different because it can operate on language, context, and unstructured data, which means it can sit closer to the work humans actually do in Salesforce, Microsoft 365, Jira, GitHub, and the rest of the modern software stack. That raises the stakes for incumbent vendors because AI does not merely automate around them; in some cases, it can sit in front of them and reduce the need to interact with them directly.
There is also a structural shift in how enterprises evaluate value. Traditional software buying often rewarded feature breadth, integration depth, and ease of rollout. AI changes the debate toward outcome density: how many hours, hires, or licenses can be eliminated for a given system? That makes pricing models more fragile, especially when vendors charge per seat but customers increasingly want per outcome or per workflow. The Information’s report is effectively a warning shot to the software industry that the old assumptions are being stress-tested in real budgets.
At the same time, tf imminent collapse. The same report also lays out the bull case for traditional enterprise software: core systems of record still matter, reliability still matters, and AI is often layered on top of existing platforms rather than replacing them. That tension defines the current moment. The market is caught between hype about AI-driven substitution and the operational reality that most companies still need governed systems, audit trails, and predictable workflows.
The result is an enterprise software market where AI is both a feature and a threat. It is a feature because vendors are racing to add copilots, agents, and search layers to their products. It is a threat because the better those AI layers become, the easier it is for customers to ask whether they need as many paid seats in the first place. That dynamic is what makes this report more than just another AI think piece.
The report’s Deep Research summary pointples of AI being used to cut costs, and the through line across those examples is immediate ROI. In one case, Valvoline used Torq AI to avoid buying additional CrowdStrike licenses. In another, companies used AI to absorb work after layoffs, which reduces the need for the software seats tied to those employees. Bonterra reportedly moved engineers off GitHub Copilot to Cursor when the newer tool was more effective, and the City of Sioux Falls dropped paid Copilot licenses in favor of free alternatives.
Still, the direction is hard to miss. Companies are not merely asking which AI model is smartest. They are asking which tool can replace paid work, reduce license counts, or let them avoid a new procurement cycle altogether. That is a much more commercial question, and it puts pressure on vendors to prove that AI features generate net savings instead of just adding another line item.
The most important takeaway is that enterprise AI is now reaching the budget line. Once that happens, the conversation changes from innovation theater to operational discipline. Buyers begin to compare AI tools against the cost of keeping the old workflow intact, and incumbents must defend themselves with measurable outcomes rather than abstract promises.
The report’s examples illustrate this pressure in a direct way. If AI can help teams avoid hiring, then the company has fewer employees to license. If AI can help existing staff do more with fewer tools, then it may not just shrink headcount growth; it may also shrink the need to expand software subscriptions in lockstep. That is the kind of compression legacy vendors dread because it comes not from one rival, but from a new operating model.
This also explains why AI can be more disruptive in some functions than others. In sales operations, support, engineering, and security workflows, employees spend much of their time moving data between tools, summarizing information, or initiating standard actions. If AI absorbs those activities, software usage can become shallower even when the underlying business remains healthy. In other words, the customer keeps operating, but the vendor gets less of the value.
Legacy vendors still have an answer, but it is not a guaranteed one. They can embed AI natively, bundle it into premium plans, and try to keep the customer inside the ecosystem. Yet that response only works if the AI is good enough to justify the price and easy enough to adopt without creating new friction. If not, customers will keep looking for cheaper, swappable options.
The Information’s report underscores this danger by highlighting examples where companies either bypassed new software purchases or switched AI tools when a better option appeared. Bonterra’s move from GitHub Copilot to Cursor is a reminder that loyalty may be lower in AI than it was in traditional enterprise software, especially when buyers can compare tools quickly and abandon one without much sunk cost. The City of Sioux Falls decision to drop paid Copilot licenses in favor of free tools is even more telling because it suggests that distribution alone will not protect pricing power.
That creates a new form of vendor risk: the product may still exist, but the user interface becomes commoditized. If a general-purpose AI agent can work across multiple systems, the underlying software loses its role as the main point of interaction. That is what makes the report’s “dumb pipes” concern so consequential. The software may still store records, but it is no longer where the work happens.
The bear case also extends to pricing psychology. Customers that see AI as interchangeable will resist premium markups unless a vendor proves unique value. They will compare feature sets quickly, use pilots aggressively, and switch more readily than they did in the pre-AI era. That means software giants can no longer rely solely on brand, history, or installed base. They have to earn each AI dollar.
That is why The Information’s analysis has resonated in the market. It does not claim enterprise software is dead. It argues something more uncomfortable: the old growth model may no longer scale the way it used to. If that proves true, software companies will need to defend not just their products, but their entire economic logic.
The report’s own framing leaves room for this view. Companies may be experimenting with AI to cut costs, but that does not mean they are abandoning their enterprise software stack. In many cases, they are trying to make the stack more efficient, not replace it. That distinction matters because a productivity boost can increase the value of a platform if it helps the customer get more out of the same system.
There is also a compliance reality that AI evangelists sometimes understate. Enterprises cannot always let an autonomous agent make decisions without oversight. They need logs, controls, data governance, and security boundaries. In many environments, those requirements favor the established vendors, especially when they can layer AI features into platforms customers already trust.
The bull case therefore depends on a simple argument: AI makes incumbent software more valuable by reducing friction, not less valuable by eliminating need. If vendors can use AI to help customers complete work faster inside the same platform, they can defend their position and even expand it. The key is whether the AI is additive or substitutive.
That does not make them safe, but it does mean disruption is likely to be uneven. Some functions will be much more exposed than others. Support, content operations, and repetitive engineering tasks may be vulnerable first. Core ERP, finance, and compliance systems may resist much longer. That unevenness gives major vendors an opening to defend the most critical parts of the stack while adapting elsewhere.
The Information’s examples suggest that customers are willing to move if the value case is strong enough. Bonterra’s shift from Copilot to Cursor is a warning to incumbents that early leadership does not guarantee lasting preference. Free alternatives replacing paid Copilot licenses show that even the biggest distribution advantages can be challenged if the customer concludes the incremental value is not worth the expense.
Incumbents, however, have the advantage of trust, distribution, and embedded data. They already sit inside the workflow and have procurement relationships that startups must spend years building. If they execute well, they can bundle AI into a wider platform and turn scale into a moat. The problem is that AI makes the boundaries between products more porous, which gives focused challengers more room to attack.
This is also where AI strategy and software economics collide. If a company’s assistant becomes the primary interface, then the assistant provider can capture value even when the underlying software is commodity-like. That means software vendors may need to defend not only their applications, but also their position as the default destination for work. Whoever owns that layer owns a large share of the economic upside.
For consumers, the effect is more indirect. A worker may benefit from better tools, faster drafting, or smarter search, but the financial consequence is usually absorbed by ther software, the market can tolerate novelty, inconsistency, and experimentation for longer. Enterprise buyers cannot. They need ROI, governance, and a cleaner explanation of why one tool deserves budget over another.
Consumer software has a different rhythm. Users can adopt or abandon tools quickly, and the product can survive on appeal, convenience, or novelty. In the enterprise, by contrast, a tool must survive procurement, legal review, and internal politics. That makes the enterprise AI market slower, but also far more consequential.
That is why the same AI feature can be celebrated in one context and rejected in another. If an employee can save time with an external tool, that may be enough for a consumer. If that same tool exposes sensitive data or creates compliance risk, the enterprise may shut it down. The tension between convenience and control is going to shape the market for years.
This is where AI becomes a budgeting weapon. If a CIO can show that an AI deployment saves money by preventing software sprawl or reducing headcount growth, the decision becomes easier to justify. If the deployment merely adds another abstraction layer without measurable savings, it will face much more resistance. The Information’s examples are valuable because they show companies already making these calculations in public.
That said, subtraction is not always straightforward. Sometimes AI creates new costs in model usage, governance, integration, or monitoring. Vendors that ignore those hidden costs will find buyers becoming skeptical quickly. The real winners will be those who can prove that the net cost is lower and the net outcome is better.
This is a difficult environment for vendors because the buying committee may not see AI as a category with an automatic premium. They will ask whether the outcome can be achieved with existing tools, whether the assistant is really needed, and whether the organization can tolerate the new risk surface. The burden of proof is rising fast.
The other major question is whether vendors can turn AI into something that customers cannot easily swap out. If they can make their AI features deeply contextual, reliable, and tied to unique enterprise data, they may p. If not, AI will keep behaving like a pressure valve, giving buyers more leverage and suppliers less certainty.
Source: The Information Software Under Siege: Enterprise AI Report Card
For most of the past decade, enterprise software enjoyed a remarkably stable playbook. Vendors sold seats, expanded through adjacent modules, and relied on the fact that growing companies tended to add more people, more processes, and more software licenses at the same time. AI disrupts that logic by attacking the one assumption the model depends on most: that every new workflow needs another human user, or at least another paid account.The Information’s analysis frames this change as an early but meaningful pivot. Rather than treating AI only as a pilot-project novelty, companies are beginning to use it to rethink existing software spend, especially where repetitive knowledge work can be automated or compressed. That matters because software budgets are rarely cut all at once; they are usually reallocated one workflow at a time, until the cumulative effect becomes impossible to ignore. The report says nine recent examples show companies deploying AI to cut costs, though the impacts are uneven and still unfolding. al context is important. Earlier waves of automation mostly replaced manual labor in factories or standardized back-office tasks through rules-based systems and robotic process automation. Generative AI is different because it can operate on language, context, and unstructured data, which means it can sit closer to the work humans actually do in Salesforce, Microsoft 365, Jira, GitHub, and the rest of the modern software stack. That raises the stakes for incumbent vendors because AI does not merely automate around them; in some cases, it can sit in front of them and reduce the need to interact with them directly.
There is also a structural shift in how enterprises evaluate value. Traditional software buying often rewarded feature breadth, integration depth, and ease of rollout. AI changes the debate toward outcome density: how many hours, hires, or licenses can be eliminated for a given system? That makes pricing models more fragile, especially when vendors charge per seat but customers increasingly want per outcome or per workflow. The Information’s report is effectively a warning shot to the software industry that the old assumptions are being stress-tested in real budgets.
At the same time, tf imminent collapse. The same report also lays out the bull case for traditional enterprise software: core systems of record still matter, reliability still matters, and AI is often layered on top of existing platforms rather than replacing them. That tension defines the current moment. The market is caught between hype about AI-driven substitution and the operational reality that most companies still need governed systems, audit trails, and predictable workflows.
Why this matters now
The timing matters because 2025 and 2026 are becoming the years when pilots must turn into spending decisions. Many CIOs spent the earlier AI cycle experimenting, but experimentation does not permanently protect incumbents. Once a company discovers that a smaller team can do the work of a larger one, the budget conversation changes fast.The result is an enterprise software market where AI is both a feature and a threat. It is a feature because vendors are racing to add copilots, agents, and search layers to their products. It is a threat because the better those AI layers become, the easier it is for customers to ask whether they need as many paid seats in the first place. That dynamic is what makes this report more than just another AI think piece.
The Core Finding
The heart of The Information’s piece is simple: companies are starting to use AI not just to do more work, but to do the same work with fewer people and fewer software purchases. That is a big distinction. It shifts AI from a productivity enhancer into a cost-control mechanism, and that is when enterprise adoption becomes much harder for legacy vendors to ignore.The report’s Deep Research summary pointples of AI being used to cut costs, and the through line across those examples is immediate ROI. In one case, Valvoline used Torq AI to avoid buying additional CrowdStrike licenses. In another, companies used AI to absorb work after layoffs, which reduces the need for the software seats tied to those employees. Bonterra reportedly moved engineers off GitHub Copilot to Cursor when the newer tool was more effective, and the City of Sioux Falls dropped paid Copilot licenses in favor of free alternatives.
What the evidence suggests
These are not broad market statistics. They are concrete case studies, which is both a strength and a limitation. The strength is that they show actual behavior, not just survey sentiment. The limitation is that they may reflect early adopter behavior more than the full enterprise market, so the trend should be read as directional rather than universal.Still, the direction is hard to miss. Companies are not merely asking which AI model is smartest. They are asking which tool can replace paid work, reduce license counts, or let them avoid a new procurement cycle altogether. That is a much more commercial question, and it puts pressure on vendors to prove that AI features generate net savings instead of just adding another line item.
The most important takeaway is that enterprise AI is now reaching the budget line. Once that happens, the conversation changes from innovation theater to operational discipline. Buyers begin to compare AI tools against the cost of keeping the old workflow intact, and incumbents must defend themselves with measurable outcomes rather than abstract promises.
- AI is moving from experimentation to budget impact.
- License avoidance is becoming a measurable use case.
- Workflow substitution is more dangerous than headline automation.
- Seat-based pricing looks increasingly exposed.
- Free or lower-cost alternatives now have a credible opening.
- Productivity gains are being judged against hard cost savings.
Seat Economics Under Pressure
Seat-based pricing has always been one of enterprise software’s most durable advantages. It is easy to understand, easy to forecast, and easy to expand as organizations hire more employees. AI breaks that elegance by changing the equation from “how many users do we have?” to “how many of those users still need the full software stack at all?” That is a very different question.The report’s examples illustrate this pressure in a direct way. If AI can help teams avoid hiring, then the company has fewer employees to license. If AI can help existing staff do more with fewer tools, then it may not just shrink headcount growth; it may also shrink the need to expand software subscriptions in lockstep. That is the kind of compression legacy vendors dread because it comes not from one rival, but from a new operating model.
The license-avoidance problem
License avoidance is especially powerful because it is invisible in traditional revenue metrics until the renewal cycle arrives. A customer does not always announce that it has made a software vendor less central; it simply stops expanding the contract. That means the damage can show up first as slower net retention, fewer upsells, or weaker seat growth, long before it appears as a headline decline.This also explains why AI can be more disruptive in some functions than others. In sales operations, support, engineering, and security workflows, employees spend much of their time moving data between tools, summarizing information, or initiating standard actions. If AI absorbs those activities, software usage can become shallower even when the underlying business remains healthy. In other words, the customer keeps operating, but the vendor gets less of the value.
Legacy vendors still have an answer, but it is not a guaranteed one. They can embed AI natively, bundle it into premium plans, and try to keep the customer inside the ecosystem. Yet that response only works if the AI is good enough to justify the price and easy enough to adopt without creating new friction. If not, customers will keep looking for cheaper, swappable options.
- Seat growth is no longer guaranteed by headcount growth.
- AI can reduce both hiring and subscription expansion.
- Renewal periods become the danger zone for incumbents.
- Bundled AI only works if it is clearly better than alternatives.
- License avoidance is often cheaper than license replacement.
- Small workflow changes can produce large revenue effects.
The Bear Case for Traditional Enterprise Software
The bear case is that AI becomes the new control layer above legacy systems. In that scenario, employees no longer open every app to complete a task. Instead, an agent gathers the data, processes it, and only touches the underlying software when necessary. That makes the application itself less visible and potentially less central, turning many incumbents into background infrastructure rather than workflow destinations.The Information’s report underscores this danger by highlighting examples where companies either bypassed new software purchases or switched AI tools when a better option appeared. Bonterra’s move from GitHub Copilot to Cursor is a reminder that loyalty may be lower in AI than it was in traditional enterprise software, especially when buyers can compare tools quickly and abandon one without much sunk cost. The City of Sioux Falls decision to drop paid Copilot licenses in favor of free tools is even more telling because it suggests that distribution alone will not protect pricing power.
Swappability is the enemy of lock-in
One of enterprise software’s traditional strengths has been switching costs. Once a company deploys a platform deeply enough, it becomes expensive and risky to replace. AI weakens that moat when users interact with the assistant rather than the app, because the assistant itself becomes the daily habit and the app becomes the data backend.That creates a new form of vendor risk: the product may still exist, but the user interface becomes commoditized. If a general-purpose AI agent can work across multiple systems, the underlying software loses its role as the main point of interaction. That is what makes the report’s “dumb pipes” concern so consequential. The software may still store records, but it is no longer where the work happens.
The bear case also extends to pricing psychology. Customers that see AI as interchangeable will resist premium markups unless a vendor proves unique value. They will compare feature sets quickly, use pilots aggressively, and switch more readily than they did in the pre-AI era. That means software giants can no longer rely solely on brand, history, or installed base. They have to earn each AI dollar.
Why investors are nervous
Investors dislike uncertainty, and AI introduces multiple kinds at once. There is product uncertainty, because it is not yet clear which AI workflows will prove durable. There is pricing uncertainty, because many vendors are still experimenting with how to charge for AI. And there is demand uncertainty, because customers may buy fewer seats, not more, if AI really does replace tasks rather than merely assist them.That is why The Information’s analysis has resonated in the market. It does not claim enterprise software is dead. It argues something more uncomfortable: the old growth model may no longer scale the way it used to. If that proves true, software companies will need to defend not just their products, but their entire economic logic.
- AI agents can sit above legacy applications.
- User loyalty may shift from software suites to assistants.
- Switching costs may fall if workflows become tool-agnostic.
- Free tools become more dangerous when they are “good enough.”
- Premium pricing needs proof of unique business impact.
- Investors now have to model slower seat expansion.
The Bull Case for Enterprise Software
The bull case is just as real, and arguably more grounded in the current enterprise environment. Core software systems are not easy to rip out, especially when they handle payroll, finance, compliance, security, or regulated customer data. In those categories, the system of record still matters, and AI often works best as an enhancement to existing infrastructure rather than a replacement for it.The report’s own framing leaves room for this view. Companies may be experimenting with AI to cut costs, but that does not mean they are abandoning their enterprise software stack. In many cases, they are trying to make the stack more efficient, not replace it. That distinction matters because a productivity boost can increase the value of a platform if it helps the customer get more out of the same system.
The system-of-record moat
The strongest incumbents still control trusted, deeply embedded data. Workday, Salesforce, SAP, Microsoft, and similar platforms are not just apps; they are operating frameworks for the business. Replacing them is expensive, operationally risky, and often politically difficult. AI may improve the interface, but it does not automatically erase the need for structured data, permissions, and auditability.There is also a compliance reality that AI evangelists sometimes understate. Enterprises cannot always let an autonomous agent make decisions without oversight. They need logs, controls, data governance, and security boundaries. In many environments, those requirements favor the established vendors, especially when they can layer AI features into platforms customers already trust.
The bull case therefore depends on a simple argument: AI makes incumbent software more valuable by reducing friction, not less valuable by eliminating need. If vendors can use AI to help customers complete work faster inside the same platform, they can defend their position and even expand it. The key is whether the AI is additive or substitutive.
Why incumbents still have time
Another reason the bull case remains credible is that enterprises move slowly. Even when a superior tool exists, adoption depends on integration, training, procurement, security review, and change management. Those friction points buy incumbents time to improve their products and bundle AI into existing contracts.That does not make them safe, but it does mean disruption is likely to be uneven. Some functions will be much more exposed than others. Support, content operations, and repetitive engineering tasks may be vulnerable first. Core ERP, finance, and compliance systems may resist much longer. That unevenness gives major vendors an opening to defend the most critical parts of the stack while adapting elsewhere.
- Systems of record still anchor enterprise operations.
- Compliance and audit needs slow disruptive replacement.
- AI often adds value inside the existing stack.
- Big vendors can bundle AI into trusted platforms.
- Enterprise adoption is slower than consumer adoption.
- Critical workflows remain harder to automate than demos suggest.
The Competitive Implications
The competitive landscape is becoming more complicated for both incumbents and startups. Large software companies are under pressure to prove that their AI features are not cosmetic, while startups are under pressure to show that they can deliver enough trust and integration to matter inside enterprise environments. That leaves the market split between scale and agility, with no easy winner.The Information’s examples suggest that customers are willing to move if the value case is strong enough. Bonterra’s shift from Copilot to Cursor is a warning to incumbents that early leadership does not guarantee lasting preference. Free alternatives replacing paid Copilot licenses show that even the biggest distribution advantages can be challenged if the customer concludes the incremental value is not worth the expense.
Startups versus giants
Startups benefit from speed and focus. They can build around a narrow problem, ship quickly, and target specific workflows with sharp product design. That is especially valuable in AI, where the underlying model improvements happen rapidly and user expectations are still forming. A startup can win by being the best tool for one job.Incumbents, however, have the advantage of trust, distribution, and embedded data. They already sit inside the workflow and have procurement relationships that startups must spend years building. If they execute well, they can bundle AI into a wider platform and turn scale into a moat. The problem is that AI makes the boundaries between products more porous, which gives focused challengers more room to attack.
Platform control becomes the real prize
The real competition is no longer just about whether a vendor has AI. Nearly everyone does. The better question is who controls the workflow layer that decides which data gets pulled, which tasks get automated, and which systems get touched. That is where the next generation of enterprise software value will likely be created or destroyed.This is also where AI strategy and software economics collide. If a company’s assistant becomes the primary interface, then the assistant provider can capture value even when the underlying software is commodity-like. That means software vendors may need to defend not only their applications, but also their position as the default destination for work. Whoever owns that layer owns a large share of the economic upside.
- The market is shifting from feature competition to workflow control.
- Startups can win narrow use cases faster than incumbents can.
- Incumbents still have distribution and trust advantages.
- The assistant layer may capture more value than the app layer.
- Platform bundles will be used aggressively to defend share.
- Integration quality is becoming a stronger moat than branding alone.
Enterprise vs Consumer Impact
The enterprise impact is immediate because budgets, headcount, and software renewals are all on the table. When AI cuts work or avoids a new license, the effect flows directly into procurement. That is why the report feels especially important for enterprise software buyers: it is about how AI changes what businesses pay for, not just what employees can do faster.For consumers, the effect is more indirect. A worker may benefit from better tools, faster drafting, or smarter search, but the financial consequence is usually absorbed by ther software, the market can tolerate novelty, inconsistency, and experimentation for longer. Enterprise buyers cannot. They need ROI, governance, and a cleaner explanation of why one tool deserves budget over another.
Why the enterprise lens matters more
In enterprise settings, AI is being evaluated as an operational asset. That means the bar is not whether the tool is impressive, but whether it changes staffing, spending, or cycle time in ways management can verify. If it does not, the tool risks being treated as a nice-to-have rather than a must-have.Consumer software has a different rhythm. Users can adopt or abandon tools quickly, and the product can survive on appeal, convenience, or novelty. In the enterprise, by contrast, a tool must survive procurement, legal review, and internal politics. That makes the enterprise AI market slower, but also far more consequential.
The human factor
There is also a cultural difference. Consumers often want AI that feels magical, while enterprises want AI that feels controllable. A consumer assistant can be playful or imperfect as long as it is useful. A business assistant must be accurate, auditable, and secure enough to fit into a company’s risk posture.That is why the same AI feature can be celebrated in one context and rejected in another. If an employee can save time with an external tool, that may be enough for a consumer. If that same tool exposes sensitive data or creates compliance risk, the enterprise may shut it down. The tension between convenience and control is going to shape the market for years.
- Enterprise buyers demand measurable ROI.
- Consumer adoption can be driven by novelty.
- Compliance risk matters far more inside companies.
- Governance is a feature, not a footnote.
- Business users need accuracy more than delight.
- Consumer and enterprise AI will diverge in pricing and controls.
The New AI Procurement Logic
What is emerging is a new procurement logic built around outcome substitution. Buyers are no longer asking only what a tool does; they are asking what it replaces. Does it replace an extra hire? A paid seat? A manual review step? A support ticket? That shift is subtle, but it changes how software gets evaluated and how budgets are approved.This is where AI becomes a budgeting weapon. If a CIO can show that an AI deployment saves money by preventing software sprawl or reducing headcount growth, the decision becomes easier to justify. If the deployment merely adds another abstraction layer without measurable savings, it will face much more resistance. The Information’s examples are valuable because they show companies already making these calculations in public.
Buying for subtraction, not addition
Traditional software buying was often additive. You bought more software because you had more people, more data, or more workflows. AI flips that script by promising subtraction: fewer steps, fewer licenses, fewer manual tasks. That is why the best AI pitch in enterprise may not be “look what this can do,” but “look what you can stop paying for.”That said, subtraction is not always straightforward. Sometimes AI creates new costs in model usage, governance, integration, or monitoring. Vendors that ignore those hidden costs will find buyers becoming skeptical quickly. The real winners will be those who can prove that the net cost is lower and the net outcome is better.
The importance of comparability
The procurement process will also become more comparative. Buyers will benchmark one AI tool against another, and against simply not buying anything at all. Free alternatives, internal scripts, and manual workflows all become valid competitors when the use case is narrow enough.This is a difficult environment for vendors because the buying committee may not see AI as a category with an automatic premium. They will ask whether the outcome can be achieved with existing tools, whether the assistant is really needed, and whether the organization can tolerate the new risk surface. The burden of proof is rising fast.
- AI purchases are increasingly judged by what they replace.
- Net savings matter more than feature count.
- Hidden costs can undermine the ROI story.
- Free or internal tools are serious competitors.
- Buyers are comparing AI to doing nothing.
- Procurement teams now want subtraction, not just innovation.
Strengths and Opportunities
The report’s strongest contribution is that it moves the enterprise AI discussion out of abstraction and into budget reality. It shows where AI is already having an effect, even if those effects are uneven and still early. For software vendors, that clarity is useful because it identifies the exact areas where value is being won or lost. For customers, it creates a more honest framework for deciding what to automate and what to keep.- AI is finally being tested against real spending decisions.
- Buyers now have proof points for license avoidance.
- Vendors can still win by embedding AI deeply into workflows.
- The system-of-record moat remains meaningful.
- Cost cutting makes AI easier to justify internally.
- Specialized tools can outperform broad incumbents in narrow tasks.
- Enterprises can use AI to reduce operational drag.
Risks and Concerns
The downside of this shift is that enterprises may move too quickly and assume AI savings are always durable. A tool that looks efficient in a pilot may be expensive to maintain at scale, especially if it introduces reliability issues or governance overhead. There is also a risk that companies will mistake short-term headcount reductions for long-term strategic advantage. The reality may be messier.- AI savings can be overstated in early deployments.
- New tools may add governance and integration costs.
- Hallucinations and errors remain operational risks.
- Vendors may overbundle AI to preserve revenue.
- Customer switching could become more chaotic.
- Security and compliance concerns may slow adoption.
- Some workflows may prove cheaper in old tools like Excel than in AI-first systems.
Looking Ahead
The next phase of this story will be decided less by hype than by operating results. Watch whether AI continues to help companies avoid new software purchases, or whether vendors adapt fast enough to keep customers inside their ecosystems. The most important signals will come from renewals, seat counts, and how often companies decide that an AI alternative is simply better value than the incumbent product. The headline risk for legacy software is not instant collapse; it is gradual erosion disguised as efficiency.The other major question is whether vendors can turn AI into something that customers cannot easily swap out. If they can make their AI features deeply contextual, reliable, and tied to unique enterprise data, they may p. If not, AI will keep behaving like a pressure valve, giving buyers more leverage and suppliers less certainty.
- Track seat growth versus headcount growth.
- Watch for more cases of license avoidance.
- Monitor whether free AI tools keep replacing paid ones.
- Follow renewals at companies adopting AI aggressively.
- Compare native AI bundles with third-party tools.
- Watch whether core systems of record remain sticky.
- Look for spending shifts from seats to outcomes.
Source: The Information Software Under Siege: Enterprise AI Report Card
Similar threads
- Replies
- 1
- Views
- 8
- Article
- Replies
- 0
- Views
- 34
- Featured
- Article
- Replies
- 0
- Views
- 8
- Replies
- 0
- Views
- 12
- Featured
- Article
- Replies
- 0
- Views
- 6