Trojan Horse in the Cloud
Last updated on: 18th June, 2026 | Student Insights | Sagnik Chowdhury | PGDM Co'27 << back to blog
Published on: 18th June, 2026
Trojan Horse in the Cloud: How Ecosystem Immersion and Frictionless Accessibility Are Quietly Dismantling Legacy Enterprise Software Moats
Microsoft executive Charles Lamanna made a striking forecast back in February of 2024. Salespeople, he argued, would go into their CRM systems "far less, maybe never in the future." Instead, Microsoft's Copilot would pull information from systems such as Salesforce, update records on their behalf, and allow them to work with familiar tools like Outlook and Teams.At first glance, this sounded like another lofty AI claim. But it captures a deeper shift now unfolding across enterprise software and one whose consequences are being felt from Silicon Valley to Bengaluru.
Companies are not abandoning Salesforce-style applications. They are increasingly engaging with them via AI assistants that sit on top of them. The software still matters. Employees simply stop opening it.What it previews is a $2 trillion question. In early 2026, software stocks suffered a brutal reprice as investors questioned whether AI would erode the economics that had powered the SaaS boom. Reuters reported that roughly $2 trillion in market value had evaporated from the sector's peak, with Atlassian, Workday, and Salesforce among the hardest hit. Atlassian fell 47%, Workday roughly one-third, and Salesforce about 30%.Markets don't move like that over a bad quarter. They move when an old assumption quietly stops being true: that software value is based on a human loading up an app and clicking around inside it. The deeper threat to enterprise software isn't that AI is smarter. It's that the application itself is becoming invisible.
A familiar pattern, repeating
This isn't the first time the underpinnings of enterprise software have shifted. Value migrated from mainframe hardware, to the PC operating system, to cloud infrastructure, and finally to the SaaS application layer where Salesforce, SAP, and Workday built durable businesses by owning specific workflows.
No shift fully swept away the layer before it. It merely diminished its command.Value is migrating again, toward what might be called the orchestration layer: whoever decides which AI agent does what, with what data, under whose authority.
For India's large IT services firms (TCS, Infosys, Wipro, HCLTech) and the country's 1,600-plus Global Capability Centres, this is the shift that defines the next decade. The question is not whether to participate, but where to stand when it lands.
The horse that already got in
The Trojan Horse in this article's title is not a metaphor for AI creeping into enterprises now. The horse arrived years ago, in tools too routine to raise procurement objections.It arrived as Office, then Outlook and Excel, mundane enough that no one treated them as strategic. Then Teams became where meetings happened, extending Microsoft's reach from documents into conversations. Azure became the infrastructure underneath much of it. Each addition felt incremental.Together, they gave Microsoft ownership of the pipes through which modern work flows.Only now does the full shape become visible.
Copilot sat on top of everything already built. And last came Entra Agent ID, which assigns verified identities to AI agents inside corporate networks.The payload did not need to be smuggled in. Each product walked through the front door, one procurement decision at a time.
Why AI changes the equation
For three decades, enterprise software moats rested on one quiet assumption: a human would log in, navigate an interface, and do the work. SAP's R/3, launched in 1992, embedded itself so deeply into how companies ran finance and manufacturing that replacing it was not a software decision but an organisational restructuring worth an estimated footprint of $10–50 million. Salesforce built a comparable moat through customer data and a certification ecosystem that created an entire workforce of administrators, consultants, and developers invested in the platform's success.AI agents don't log in. They don't need a username or a dashboard. An agent reads a database, calls an API, and produces an outcome (a drafted email, an updated record, a completed approval) without anyone touching a screen.
The old mode was: Human → Interface → Application → Outcome. The new model is: Human → Agent
→ Application (in background) → Outcome. A 2026 Retool survey of 817 enterprise software builders found that 35% had already replaced at least one SaaS tool with a homegrown alternative, and 78% planned to build more that year.
Applications that were once destinations are increasingly evolving into infrastructure, quiet repositories that agents consult on a human's behalf. Salesforce becomes a customer database an agent queries. SAP becomes a repository of processes an agent executes against. Workday becomes an employee record an agent reads from. The software has not become less important. It has become less visited. The user moves up one layer, to the agent, while the application recedes into the background.
Microsoft's real advantage isn't AI
It is tempting to read Microsoft's position as an AI story. Copilot is genuinely capable: Microsoft 365 serves more than 400 million commercial users, and Microsoft says more than 60% of Fortune 500 companies had adopted Copilot by early 2024. But Copilot is the payoff, not the advantage.The real asset is distribution. Microsoft does not need to persuade employees to adopt a new tool, because employees already live inside Word, Excel, Outlook, and Teams every working hour. Copilot is sold as an upgrade, roughly $30 per user per month, to something already paid for and already open on every screen.
Companies are already beginning to operationalize this shift. Schneider Electric, for example, deployed Microsoft Copilot for Sales to connect Outlook, Teams, and Salesforce, allowing sellers to access customer insights, meeting summaries, and CRM actions directly within their existing workflow. As noted, the goal was to bring AI capabilities "directly in the workflow of salespeople" rather than requiring constant movement between applications.
And every Copilot interaction feeds the Microsoft Graph: a running map of how work happens inside a company, who communicated with whom, which documents were edited, which meetings produced which decisions. No competitor has an equivalent asset without access to that same layer of daily work. With Entra Agent ID, Microsoft is extending that position into the identity and permissions layer that AI agents increasingly rely on. The horse didn't need to be clever. It needed to already be standing in the courtyard when the gates opened.
Google's different bet
Google's response makes the contrast clear. Microsoft owns the workflow and is adding AI to it. Google owns the AI and is trying to access the workflow.In October 2025, Google merged its enterprise tools into Gemini Enterprise, and the headline feature wasn't a smarter model. It was connectors letting Gemini read directly from Microsoft 365, SharePoint, Salesforce, and SAP. Google's own announcement was blunt: an agent is only as good as its context, and that context lives "from Google Workspace and Microsoft 365 to business applications like Salesforce and SAP."
In plain terms, Google doesn't need your company to leave Microsoft. It needs to become the smartest layer of intelligence sitting on top of Microsoft's data, winning the intelligence war by reading its competitor's mail (with consent, of course).
Who's most exposed
Not every incumbent faces equal risk. SAP's core finance and operations software is difficult to displace in the near term as no CFO replaces a general ledger because a chatbot exists. But SAP's adjacent products, including its CRM, analytics, and spend-management tools, are in real jeopardy, without that same depth of lock-in.Salesforce faces a harder problem. Its new product, Agentforce, charges $2 per "conversation" instead of per seat. To match the revenue of one $150-a-year seat, Salesforce needs roughly 75 agent conversations per user annually, from precisely the high-activity users whose work agents are now automating away. ServiceNow has taken a cleverer path, repositioning itself not as a tool agents replace but as the platform agents need to fit into.
AWS has the opposite problem from Microsoft. It holds approximately 30% of the global cloud market but has no email client, no word processor, and no collaboration layer. It owns the basement, not the first floor. Its April 2026 launch of a productivity tool called Amazon Quick was an acknowledgement of exactly that gap.
India's position in the same shift
This orchestration shift is the same one India's IT services industry is navigating, and the stakes run in both directions. TCS, Infosys, Wipro, and HCLTech sit at an unusual crossroads: they are both implementers of this transformation and, in their traditional lines of business, exposed to it. Infosys reported deploying over 500 AI agents across client sites in early FY2026 and credited its enterprise AI positioning for industry-leading growth during that period. India's 1,600-plus GCCs, employing 1.8 million professionals, are evolving from back-office processing hubs into AI orchestration centres, designing and governing the agents that are reshaping enterprise software from the inside.
The risk is just as real. Routine, rules-based work (application maintenance, ticket processing, data entry) has been the foundation of India's IT services revenue for two decades. This is exactly the category agents automate first. Firms and professionals who move toward agent design and AI governance in the next 18 to 24 months will be better placed to shape what this layer becomes. The window is narrower than it appears.
Why the moat may not be moving
It would be too neat to declare enterprise software moats dead, and for a moment, the opposite case feels almost more convincing. As foundation models commoditize, scarcity shifts to data. Salesforce's Data Cloud, SAP's ten year cliff of transaction histories, and Oracle's databases full of mission-critical records don't lose value in this world, they gain it, becoming the raw material every agent relies on. Under this reading, incumbents aren't being disrupted at all. As fewer and fewer people actually see them directly, they're nevertheless becoming a little bit more crucial under the radar.
But notice what that argument concedes. It doesn't dispute that applications are becoming invisible, it argues that invisibility is a power position for the holder of the source data. Both can be true at once, and that's the deeper point: You don't have to kill the application for its moat to move. It just needs to stop being where the relationship with the user exists.
There's also genuine friction in execution. Gartner predicts 40% of agentic AI projects will be axed by 2027 over unclear value and weak governance. Klarna's well-known 2024 experiment, cancelling Salesforce and Workday contracts, replacing 700 staff with AI, then rehiring humans in 2025 after quality dropped, is a caution in both directions. The disruption is real, but so is the challenge of getting it right.
The idea worth remembering
The most important change underway is not that software is disappearing. It is that software is becoming invisible. Employees do not need to stop using Salesforce for Salesforce to become less central to their daily work. They only need to stop opening it. As agents increasingly mediate how people interact with information, the relationship between users and applications begins to loosen — even when the applications themselves remain indispensable.
References
- Charles Lamanna CRM/Copilot quote ("far less, maybe never in the future")
Reuters: https://www.reuters.com/technology/microsoft-tailors-ai-finance-teams-specialized-copilot-2024-02-29/ - Software sector AI selloff (~$2 trillion market value loss)
Reuters: https://www.reuters.com/business/software-real-estate-us-sectors-under-grip-ai-scare-trade-2026-02-13/
Wall Street Journal: https://www.wsj.com/finance/stocks/the-1-6-trillion-meltdown-that-swept-through-software-stocks-86c8b3a2 - Microsoft Entra Agent ID
https://techcommunity.microsoft.com/blog/microsoft-entra-blog/announcing-microsoft-entra-agent-id-secure-and-manage-your-ai-agents/3827392 - SAP R/3 and enterprise ERP adoption
https://www.sap.com/about/company/history/1991-2000.html
https://community.sap.com/t5/enterprise-resource-planning-blog-posts-by-members/a-brief-history-of-sap-an-introduction-for-the-uninitiated/ba-p/13885008 - Salesforce certification ecosystem
https://trailhead.salesforce.com/credentials/verification
https://trailhead.salesforce.com/content/learn/modules/salesforce-credentials-quick-look/get-familiar-with-salesforce-credentials-on-trailhead - Retool Build vs Buy Report 2026
https://www.businesswire.com/news/home/20260217548274/en/Retools-2026-Build-vs.-Buy-Report-Reveals-35-of-Enterprises-Have-Already-Replaced-SaaS-With-Custom-Software - Microsoft 365 Copilot adoption
Microsoft Ignite 2024: https://blogs.microsoft.com/blog/2024/11/19/ignite-2024-why-nearly-70-of-the-fortune-500-now-use-microsoft-365-copilot/
Microsoft SEC Filing: https://www.sec.gov/Archives/edgar/data/789019/000119312524242888/d815777dars.pdf - Microsoft Copilot pricing
https://www.microsoft.com/en-us/microsoft-365/copilot/business - Google Gemini Enterprise
https://cloud.google.com/blog/products/ai-machine-learning/introducing-gemini-enterprise - Salesforce Agentforce pricing
https://www.salesforce.com/news/press-releases/2025/05/15/agentforce-flexible-pricing-news/ - Global cloud market share (AWS ~28–31%)
Synergy Research: https://www.srgresearch.com/articles/cloud-market-share-trends-big-three-together-hold-63-while-oracle-and-the-neoclouds-inch-higher
Statista: https://www.statista.com/chart/18819/worldwide-market-share-of-leading-cloud-infrastructure-service-providers/ - Infosys AI agents
Infosys SEC Filing: https://www.sec.gov/Archives/edgar/data/1067491/000106749126000018/exv99w05.htm
Reuters – Infosys ties up with Anthropic, days after IT selloff
Moneycontrol – Infosys counts over 500 AI agents, sees AI work across 90% of top 200 clients - Gartner prediction: 40% of agentic AI projects cancelled by 2027
https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027 - Klarna AI experiment
https://www.fastcompany.com/91468582/klarna-tried-to-replace-its-workforce-with-ai
https://www.salesforceben.com/klarna-cut-ties-with-salesforce-to-go-all-in-on-ai-now-theyre-hiring-humans-back/ - Amazon Quick
https://www.aboutamazon.com/news/aws/amazon-quick-desktop-ai-assistant - Schneider Electric Copilot for Sales case study
https://www.microsoft.com/en/customers/story/18860-schneider-electric-microsoft-365-copilot-for-sales - Microsoft FY24/Q2 earnings discussion (Over 400M)
https://www.microsoft.com/en-us/investor/events/fy-2024/earnings-fy-2024-q2