A key question in Bank of America’s latest software industry report is whether software vendors can lift AI revenue from low single digits into double digits.
In the report dated Oct. 20, 2026, BofA said agentic AI will not turn software into a commodity. Gartner estimates application software spending will grow at a 13.2% compound annual rate from 2025 to 2030. After factoring in cannibalization from cross-functional AI agents, that growth rate falls to 10.9%, a difference of 2.3 percentage points. BofA’s view is that lower coding costs do not remove the cost and complexity of production deployment, integration, governance and maintenance. In that setup, value shifts away from the feature layer and toward execution and governance.
Lower coding costs do not erase enterprise software complexity
The report pointed to REA, short for Reverse Engineer Anything, a tool that appeared in early October. It connects AI coding agents to Ghidra, Hopper and IDA, breaks installed software into simplified blueprints through reverse engineering, and then uses AI to recreate functionality. ArtCraft took a similar route and released seven open-source applications modeled on Photoshop, Illustrator, Premiere Pro and Lightroom, with interfaces and tool layouts close to Adobe products.
BofA said those demos amplify the narrative that anyone can build software. Replicating menus, panels and basic functions is much easier than matching mature commercial software on depth, reliability, performance and file compatibility. ArtCraft itself labeled several of its applications as alpha or in development.
The report cited a similar pattern in computer-aided design, or CAD. Open-source SolidCraft offers a parametric modeling workflow, but simulation, advanced surfacing, computer-aided manufacturing, or CAM, and 2D drawings are still unfinished.
BofA grouped agentic coding into four categories:
- prompt-only application generation
- conversational development
- AI-accelerated coding
- specification-driven development
These tools turn natural language into a software creation interface, but enterprise deployment still depends on experienced engineers, product judgment and domain expertise. The bank said AI agents are strong at replicating, adapting and iterating on known designs, while genuinely novel concepts remain harder to produce.
Moats are moving from interface design to trusted data, domain knowledge and control layers
BofA argued that UI and UX are becoming weaker moats. Agentic AI interacts with applications more through APIs, and agents can operate clunky legacy interfaces on behalf of users. That reduces the importance of interface familiarity and design quality.
The report placed the new defensive focus on four areas: trusted business data, embedded industry knowledge, AI control layers and partner networks.
On data, the bank said the issue is not just volume but how the data is created. Customer-imported data is portable. Approval histories, anomalies, escalation paths, human overrides and final outcomes generated inside live workflows are much harder to copy. ServiceNow, for example, highlights 85 billion production workflows and 7 trillion transactions as business context for AI. SAP generates similar context in core finance and supply chain processes. BofA said the quality of business context matters more than raw data volume.
AI control layers are another emerging moat. SAP requires AI agents built on third-party models to access systems through its own governed platform. ServiceNow’s AI Control Tower can discover, monitor and govern AI agents across the enterprise and can shut down agents that exceed their authority.
BofA does not expect one company to control every agent. Instead, it sees a layered structure under enterprise control towers: Microsoft for productivity tools, SAP for enterprise resource planning, or ERP, ServiceNow for IT and workflows, and communications-platform-as-a-service, or CPaaS, vendors such as Twilio and Sinch for customer communications.
The report also said system integrators and value-added reseller networks are underappreciated. Enterprise deployments rarely end when software is sold. Integration, migration, process redesign, localization, training and ongoing optimization still require partners. Agentic AI may automate routine configuration and coding, but it also creates new demand around data preparation, governance, control, audit and cross-system coordination.
Seat-based pricing is under pressure as vendors test consumption and outcome models
BofA said seat-based pricing faces a direct challenge from agentic AI. If agents do more of the work while user counts do not rise, the headcount base behind per-seat charging gets squeezed.
Software vendors are responding by testing consumption- and outcome-based pricing, charging by token, action, conversation or completed result. The report said pricing is still in an experimental phase, and customer resistance could slow the move away from subscription structures.
BofA placed SAP in a defensive position. Its AI monetization model combines seat-based and consumption pricing. Joule Copilot and Joule Studio support about 40 domain agents, and customers can combine and customize workflows. SAP disclosed that more than two-thirds of cloud orders include AI, 90% of its top 50 deals include Business Data Cloud, or BDC, or AI, fewer than 60% of cloud customers have already used AI, and $2 billion in BDC contracts has been signed. BofA rates SAP Buy with a price target of 226 euros.
CAD software is more exposed to seat-based pressure. BofA said Autodesk, Dassault Systèmes and Nemetschek derive their moat from accurate digital records across design, testing, manufacturing and product lifecycle stages. AI-generated geometry still has to satisfy engineering constraints, pass simulation and connect to downstream processes. Because all three still rely mainly on seat-based pricing, pressure on seat counts would hit their revenue models more directly.
BofA rates Autodesk Buy with a $300 price target, Dassault Systèmes Neutral with a 25.6 euro target, and Nemetschek Buy with a 93 euro target.
Gartner’s model shows a 2.3-point drag from AI cannibalization
Gartner projects a 13.2% compound annual growth rate for application software spending from 2025 to 2030. After including cannibalization from cross-functional agent applications, growth in traditional functional markets drops to 10.9%, leaving a 2.3 percentage point gap.
According to figures cited by BofA, about $116 billion of spending will shift from traditional functional categories to cross-functional agent applications from 2026 to 2030. The annual cannibalization amount rises from $400 million in 2026 to $2 billion in 2027, $11 billion in 2028, $25 billion in 2029 and $79 billion in 2030.
BofA said the 2.3-point difference reflects spending moving between categories rather than disappearing in equal measure from the total pool. Large platforms that can offer both core applications and cross-enterprise orchestration may be able to recapture that spending across categories. The risk is more concentrated among narrow vendors focused on a single function.
By segment, customer relationship management, or CRM, growth falls from 14% to 11%, ERP from 12% to 9.5%, supply chain management, or SCM, from 13.5% to 12.5%, and content services from 10% to 6%. Cross-functional agent applications are expected to account for more than 10% of application software spending, with value concentrated in vendors tied to orchestration control, enterprise data foundations and workflow execution.
The report also said AI investment decisions are increasingly shared with business units. An International Data Corporation, or IDC, survey found that 73% of agentic AI investment decisions are jointly owned by business and IT or led by the business side. Some AI spending may come from line-of-business productivity budgets, easing substitution pressure on centralized IT software budgets.
Stock calls diverge based on where moats sit
BofA put SAP in the defensive camp. ERP spans finance, human resources, supply chain, procurement and customer experience. Deep data models, embedded controls and regulated workflows raise switching costs, making replacement by general-purpose agents harder. The report said SAP’s risk lies in edge cases where generative AI can substitute, but regulatory requirements, mission-critical coverage and switching costs still tilt the balance toward SAP.
In travel IT, BofA said the top of the distribution funnel will be reshaped by personal agents. It rates Amadeus Underperform with a 53 euro target, Sabre Buy with a $52.5 target, and HBX Buy with a 12 euro target. At the same time, the bank said personal agents are unlikely to become a mainstream booking channel soon because suppliers control inventory, user trust is limited and privacy constraints remain in place. Agents can still change how travelers discover and compare options without executing the transaction themselves.
In CPaaS, BofA said agent-driven communications should lift long-term volume growth. It rates Twilio Buy with a $320 target and Bandwidth Neutral with a $60 target.
Data platforms and observability software were described as beneficiaries of rising demand for governed data and more complex workloads created by agents. BofA rates Snowflake Buy with a $470 target, MongoDB Buy with a $540 target, Datadog Buy with a $305 target, Dynatrace Buy with a $65 target, and Elastic Neutral with a $108 target.
The split is sharper in developer tools and robotic process automation, or RPA. BofA rates JFrog Buy with a $110 target and GitLab Neutral with a $54 target. The report said AI is changing the software development lifecycle, raising complexity and security requirements and increasing the strategic value of developer tools.
For RPA, BofA rates UiPath Underperform with a $15 target. The bank said probabilistic AI agents are competing with deterministic RPA for future workload share, and improvements in AI capability or lower AI costs could erode the value proposition of RPA.
The article was presented by Chaoxiang Research as a summary and interpretation of a third-party brokerage report from Bank of America dated Oct. 20, 2026, combined with public market information. It said the ratings, price targets, earnings forecasts and related judgments cited in the piece are the views of the brokerage analysts and represent only their institution’s position, not the view of Chaoxiang Research, and do not constitute investment advice.
The piece also said market risk remains and decisions should be made independently. It should not be used as the basis for buying or selling any security.
