The pitch has been made a thousand times in Silicon Valley boardrooms: artificial intelligence will transform the way companies operate. But something different is happening now. AI agents — autonomous software programs that can reason, plan, and execute multi-step tasks without constant human oversight — aren’t a future promise anymore. They’re already driving measurable revenue at some of the world’s largest technology companies. And the financial implications are becoming impossible to ignore.
According to Yahoo Finance, AI agents are already responsible for roughly 10% of code generated at major enterprises, a figure that’s climbing fast. That statistic alone should make every CIO, CTO, and portfolio manager sit up. If autonomous AI systems can write a tenth of all production code today, the trajectory points toward a fundamental restructuring of how software gets built, tested, deployed, and maintained.
This isn’t about chatbots answering customer service questions. Not anymore.
The current generation of AI agents represents a qualitative leap from the large language models that captured public attention in 2023. Where ChatGPT and its competitors respond to single prompts, agents chain together multiple reasoning steps, access external tools, query databases, write and execute code, and even correct their own mistakes mid-task. They operate more like junior employees than search engines. Microsoft, Google, Salesforce, and ServiceNow have all introduced agent frameworks in recent months, each betting that autonomous AI will become the primary interface between humans and enterprise software within the next few years.
Microsoft’s Copilot platform has been the most visible example. CEO Satya Nadella has repeatedly emphasized that the company’s AI strategy is shifting from copilots — tools that assist humans — to agents that can operate independently. During Microsoft’s most recent earnings call, the company disclosed that its AI business had reached an annualized revenue run rate exceeding $13 billion, a figure growing more than 175% year-over-year. A significant and increasing share of that growth is coming from agent-based capabilities embedded in Microsoft 365, Dynamics, and Azure.
Google isn’t standing still. At its Cloud Next conference earlier this year, the company unveiled its Agent Development Kit and expanded the capabilities of Gemini models specifically for agentic workflows. Google Cloud CEO Thomas Kurian told attendees that enterprise customers are moving beyond experimentation and into production deployments of AI agents for supply chain management, financial analysis, and software engineering. Alphabet’s cloud division posted $12.26 billion in revenue last quarter, with AI products cited as a primary growth driver.
Salesforce has arguably been the most aggressive in branding its agent strategy. CEO Marc Benioff has positioned the company’s Agentforce platform as the next evolution of CRM, claiming it can autonomously handle customer interactions, generate sales strategies, and manage workflows across departments. During Salesforce’s Q4 fiscal 2025 earnings call, Benioff said Agentforce had closed more than 5,000 deals since its launch. The stock responded accordingly.
But here’s where it gets complicated for investors.
The revenue contribution from AI agents is real, but parsing it from broader cloud and software spending remains difficult. When Microsoft reports its AI run rate, that number bundles together copilot subscriptions, Azure AI infrastructure consumption, and agent-specific workloads. When Salesforce touts Agentforce deals, it’s not always clear how much incremental revenue those contracts represent versus standard platform renewals with an AI upsell attached. Wall Street analysts are still building the frameworks to properly value these businesses, and the lack of standardized disclosure makes comparisons across companies unreliable.
Morgan Stanley published a note in late May 2025 estimating that AI agents could add $150 billion in incremental software revenue industrywide by 2028. The firm’s logic: agents don’t just automate existing tasks — they create new categories of software consumption. A company that previously needed 50 developers might still employ 45, but those 45 developers, augmented by agents, could produce the output of 80. The net effect is more software consumed, not less. More compute. More API calls. More data storage. The infrastructure layer benefits enormously.
That thesis explains why Nvidia’s stock continues to trade at elevated multiples despite concerns about AI spending sustainability. Every autonomous agent running in production requires inference compute — the processing power needed to generate responses and take actions in real time. Unlike training, which happens once or periodically, inference is continuous. An agent handling customer support tickets runs 24 hours a day, seven days a week. Multiply that across thousands of enterprise deployments and the demand curve for GPU capacity becomes structurally different from previous technology cycles.
ServiceNow has emerged as a less obvious but increasingly important player. The company’s Now Assist platform integrates AI agents directly into IT service management, HR workflows, and security operations. CEO Bill McDermott told analysts in April that AI was present in more than 85% of new deals, with average contract values rising as customers adopt agent capabilities. ServiceNow’s positioning is particularly interesting because it sits at the intersection of workflow automation and AI — exactly where agents deliver the most immediate value.
The enterprise adoption pattern is following a predictable curve. First, companies deploy agents for internal use cases with limited blast radius: code generation, IT helpdesk automation, document summarization. Then they expand to customer-facing applications. Then — and this is where the real economic impact begins — they start redesigning entire business processes around the assumption that agents will handle significant portions of the work.
JPMorgan Chase disclosed earlier this year that it has more than 200 AI use cases in production, with agents handling tasks ranging from fraud detection to marketing copy generation. The bank estimated these deployments are saving hundreds of millions of dollars annually. That’s not a pilot program. That’s operational transformation at scale.
Not everyone is convinced the current pace is sustainable. Prominent AI researcher Gary Marcus has warned that agent reliability remains a serious concern, noting that autonomous systems making errors at even a 2-3% rate can cause significant damage when operating without human supervision. Enterprise IT leaders echo this worry privately: the technology works impressively well 95% of the time, but that remaining 5% can produce hallucinated data, incorrect code, or unauthorized actions that create liability.
The guardrail problem is real. And it’s creating a secondary market opportunity. Startups like Patronus AI, Galileo, and Arize are building monitoring and evaluation platforms specifically designed to track agent behavior, flag anomalies, and enforce policy compliance. Venture capital firms poured more than $2 billion into AI safety and observability companies in the first half of 2025 alone, according to PitchBook data.
There’s also the labor market question that no one in corporate America wants to address directly. If AI agents can generate 10% of code today and that figure reaches 30-40% within two years — a range multiple executives have privately suggested is realistic — what happens to hiring plans for software engineers? The immediate answer from most companies is that developers will shift to higher-value work: architecture, system design, complex problem-solving. But the math doesn’t always support maintaining current headcounts when productivity per engineer rises dramatically.
Klarna offered an early preview of this dynamic. The fintech company disclosed that its AI assistant was doing the work equivalent of 700 full-time customer service agents within months of deployment. Klarna subsequently reduced its workforce and cited AI as a factor. Other companies are watching closely but communicating more cautiously, aware that announcing AI-driven layoffs carries reputational and regulatory risk.
The regulatory environment is evolving in parallel. The European Union’s AI Act, which began phased enforcement in 2025, includes provisions specifically addressing autonomous AI systems. High-risk applications — those used in employment decisions, credit scoring, or critical infrastructure — face stringent transparency and human oversight requirements. U.S. regulation remains more fragmented, with state-level initiatives in California, Colorado, and New York creating a patchwork of compliance obligations that enterprise legal teams are scrambling to interpret.
So where does this leave investors trying to size the opportunity?
The most honest answer: it’s early, the TAM estimates are wide-ranging, and the competitive dynamics are far from settled. But several things seem clear. First, AI agent adoption is accelerating faster than most forecasts predicted even 12 months ago. The 10% code generation figure reported by Yahoo Finance is likely conservative given the pace of improvement in underlying models. Second, the revenue impact is showing up in earnings — not uniformly, and not always transparently, but it’s there. Third, the infrastructure demands of agentic AI are structurally bullish for cloud providers and semiconductor companies.
And fourth — perhaps most importantly — the companies that figure out how to make agents reliable enough for mission-critical applications will capture disproportionate value. Trust is the bottleneck, not capability. The technology can already do remarkable things. The question is whether enterprises will let it.
The next two earnings cycles will be telling. Investors should watch for three specific signals: the percentage of total revenue that companies explicitly attribute to AI agent products, the trajectory of average contract values in enterprise software deals, and any disclosures around AI-driven workforce restructuring. Those data points, taken together, will reveal whether the current enthusiasm is justified or whether the market has once again gotten ahead of the technology.
Right now, the evidence tilts toward justified. But barely.
WebProNews is an iEntry Publication