The launch of a new AI automation tool from Anthropic recently sent markets into a spiral, but this may say more about investor sentiment than the technology itself. While the stock selloff across software companies, financial data providers, and broader markets may have been triggered by the AI product launch, it seems that although enterprises are certainly pouring money into AI at record levels, a significant majority, potentially up to 95% of AI initiatives, still show no measurable return, according to a report from MIT.
Generative AI has been grabbing headlines, and valuations have reached sky-high levels, however, this recent investor activity has once again put agentic AI under close scrutiny, as the market seeks evidence of tangible, scalable use cases.
The risks to software from the rise of agentic AI could include increased competition, downward pricing pressure, and a potential move toward more in-house software development. Disintermediation is also a possible threat, referring to a potential shift in customers prioritising access via AI over vendor-specific software.
However, the prevailing view in the industry is that these combined risks may not result in the replacement of software by AI until implementation and proof of ROI reach far higher levels.
The financial sector is one of the earliest adopters of agentic AI. Its high-volume, compliance-heavy operations might offer the grounds for clear efficiency gains and measurable ROI. According to IDC research referenced by Statista, the financial sector will spend $97 billion on AI in 2027, while global AI spending is expected to reach $632 billion in 2028, with a 29% compound annual growth rate, and financial services accounting for around 20% of total AI spending.
Manufacturing, retail, and logistics industries are also positioned as natural early AI adopters due to high transactional volumes and measurable operational leverage in structured workflows.
The biggest challenge: Proving ROI
Despite the large sums being invested in AI, our discussions with those in the industry point to a common theme: the biggest challenge appears to be proving real value. A report from MIT found that the vast majority of enterprise AI initiatives deliver no measurable ROI at all.
Caution is still evident as monetisation is still largely unproven, and competition is likely to erode pricing power. This issue is compounded by ticket sizes that are roughly ten times higher than previous robotic process automation (RPA) solutions. Looking into 2026, most expect monetisation to remain difficult and to plateau, even though agentic capabilities may still be necessary for competitive differentiation.
While many companies are willing to explore agentic AI solutions, CFO sign-off remains limited. Experts observe that only around 20-30% of trials are moving to full implementation, and momentum is beginning to slow.
Agentic AI’s first big payoff may be process automation
Agentic process automation (APA) is the most likely next “win” for agentic AI. APA, where AI agents autonomously execute end-to-end business processes, such as invoice processing, customer onboarding, IT helpdesk support, and claims handling. APA offers clear and measurable ROI through cost reduction, faster processing, and improved compliance.
It also builds on existing RPA infrastructure, making adoption easier for enterprises that already use automation tools. This makes APA a more realistic and immediate value driver than other agentic AI use cases that are harder to measure or riskier to deploy.
However, more complex applications, such as customer experience business process outsourcing (CxBPO), have struggled to prove the value of automated Agentic AI solutions, with some players re-prioritising the need for human involvement.
APA has also led to the emergence of a broader software ecosystem, allowing process mining and workflow-related companies to capture spend. While they may not offer automation solutions directly, these ecosystem tools help identify processes suitable for automation and simplify workflow creation. This enables less technologically skilled personnel to benefit from APA.
Agentic AI growth is coming, but incumbent RPA vendors may not be winners
Investors appear to remain skeptical about existing automation players’ ability to capture market share. UiPath, for example, announced strategic partnerships with Nvidia and OpenAI that initially sparked sharp stock-market gains, with shares rising more than 20% around the announcements in late September 2025, but much of that upside later retraced as enthusiasm faded and questions about near-term monetisation persisted. Shares have lost about a third of their value year-to-date as of 26 February 2026.
An expert we spoke with said UiPath and other incumbents are prioritising the conversion of their existing RPA customer base to secure revenue during this shift, and have consequently captured about 20% of current APA spend. Competition is intensifying, and incumbents are likely to face slower expansion, with growth moderating compared with earlier years.
Taking share are new automation entrants, including hyperscalers (Amazon/AWS, Microsoft/Azure and Google/Google Cloud Platform) and independent software vendors (ISVs) such as SAP, Salesforce and ServiceNow, with both buckets capturing 40% of spend each. They are committing billions of dollars in capital expenditures to enable agentic functionality. Private markets are also backing companies with billion-dollar valuations, such as n8n.
Hyperscalers have lowered the costs of agent creation, quoting prices up to 37% cheaper than incumbents, with consumption-based spend difficult to displace later.
However, it may be the ISVs that may capture the most spend going forward and potentially pose the greatest threat to incumbents. By focusing on business processes and offering simpler, pre-packaged agent frameworks, ISV solutions such as SAP’s Joule are already displacing existing automation solutions.
Automation pure-players, including incumbents, may be able to claw some of this back by prioritising customisation; though this risks complicating customer IT infrastructure.
The slow death of software? Not yet.
The spending numbers tell one story, the implementation rates tell another. Most enterprises appear to still be figuring out where agentic AI actually fits into their operations, and CFOs are in no rush to find out the hard way.
Other barriers to AI takeover include deeply integrated software: Tier 1 vendors, including SAP, Oracle and Microsoft, benefit from cloud-native architecture, deep workflow integration and regulatory complexity, making threat of AI entrants minimal.
The overarching sentiment from the industry is that AI will not replace software completely. There are currently too many examples of significant gaps between AI expectations and actual performance. Managing AI hallucinations and failure to achieve the required quality of output have put a serious dent into the success rates for AI projects.
Early generative AI implementations have been characterized by overpromising. Vendors promised capabilities that platforms, at their level of maturity, simply could not deliver for complex business environments.
Companies also seem unwilling to fully embrace in-house solutions given the need to build out teams, meet ongoing maintenance requirements to keep software functioning, and satisfy regulatory compliance. However, the risks around disintermediation, ancillary product in-housing, increased competition, and downward pricing pressure remain firmly in play for software companies. And AI certainly isn’t going anywhere.
(Third Bridge is a primary research business, this article is based on conversations and insights with relevant experts in the industry.)
