You are reading an earlier report. A newer report on this company was published on May 20, 2026: AI Content Copyright and Data Licensing: Investment Research
LeadAgents are moving from "answering" to "acting"—becoming the enterprise system-of-action execution layer. The clearest revenue evidence on the ground is concentrated in Salesforce Agentforce (ARR 800 million, +169%), Microsoft 365 Copilot (20 million paid seats), NICE AI ARR +66%, HubSpot outcome-based pricing, Palantir Q1 +85%, and UiPath ARR 1.853 billion. The hardest layer to displace is not the model itself but permissions, the workflow entry point, connectors, governance, and distribution. Key names to track: Microsoft / Salesforce / ServiceNow / NICE / HubSpot / Palantir / UiPath / Pegasystems / Appian / SAP. Rating Watch: revenue proof is concentrating in a handful of platform and application-layer leaders, while adoption pace—not model capability—remains the swing factor for the rest.
Core Conclusions
The place AI Agents occupy in the value chain is not "a model that chats better," but the assembly of base models, enterprise data, tool invocation, permission systems, approval flows, monitoring, and auditing into a deliverable "system of action" execution layer. The durable profit pool does not sit in single-shot inference itself; it sits with whoever controls the workflow entry point, permission management, enterprise data context, and the connector network.
The fundamental difference between Agents and Copilots, chatbots, and RPA is this: Copilots mainly raise human productivity, bots/RPA mainly execute business along fixed scripts, while an Agent can autonomously plan under goal constraints, call tools across multiple steps, retain state, handle exceptions, and enter a human-in-the-loop when needed. It moves from "answering" to "acting."
Enterprises do not just buy model APIs directly, and the reason is no mystery: a model API solves "can it reason," whereas what enterprises actually need to solve is "can it connect to systems within permissions, call tools, audit behavior, control costs, trace failures, satisfy compliance, and keep improving." That is why MCP, connectors, identity, governance, observability, and the workflow layer have rapidly become table stakes.
The scenarios showing commercial validation first are not "big, do-everything general Agents," but settings with clear workflows and easily quantified ROI: AI coding, customer service / contact centers, CRM sales processes, ITSM / enterprise service desks, knowledge retrieval and enterprise search, plus HR / finance / legal subprocesses with strong rule boundaries.
As of now, the public companies with the "clearest Agent revenue on the ground" in public materials are few, and the group with the strongest evidence is Salesforce, Microsoft, NICE, HubSpot, Palantir, UiPath, and Twilio. Salesforce has disclosed that Agentforce ARR has reached 800 million USD with 169% year-over-year growth and 29,000 cumulative Agentforce deals closed. Microsoft has disclosed that Microsoft 365 Copilot paid seats now exceed 20 million, with ARPU growth once again driven by Copilot and E5. NICE has disclosed AI ARR growing 66% year over year. HubSpot has shifted its Customer Agent and Prospecting Agent to billing by "completed outcome." Palantir's Q1 2026 revenue grew 85% year over year, with US commercial revenue up 133%. UiPath ARR reached 1.853 billion USD. Twilio's Q1 2026 revenue grew 20% year over year, and its earnings materials explicitly cite AI Agents for customer service and sales processes.
Another group of large companies looks more like "defensive Agent beneficiaries" than near-term new growth engines: Oracle, SAP, Workday, and parts of Atlassian. Their greatest value is not selling Agents standalone at high ARPU, but using Agents to reinforce their existing systems of record—lifting retention, expansion, cloud backlog / RPO, and platform stickiness. Oracle has made AI Agent Studio available to Fusion Applications customers "at no additional charge." SAP's Joule already spans 35 solutions but has not separately disclosed Agent revenue. Workday is advancing around an "Agent System of Record" and role-based Agents, emphasizing managing an entire agent fleet on the customer's behalf.
On the moat, the layer most likely to form a durable barrier is not the model layer but "permission-aware data access + enterprise workflow entry point + cross-system connectors + audit and governance + a distribution ecosystem." The model layer, under multi-model competition, open source, and falling prices, more easily trends toward intensifying competition. What is genuinely hard to displace is an enterprise's organizational structure, approval logic, historical process state, and security policy.
Agents will significantly reshape software pricing models, but not as an "overnight, complete shift from seat to usage." The more realistic path is: a seat + credits hybrid model appears first, then expands in high-ROI scenarios toward charging by task, by outcome, by automated process, and by Agent workload. HubSpot already charges by resolution and recommended lead. Intercom Fin uses each resolution as the billing unit. Copilot Studio bills in Copilot Credits. Atlassian Rovo Dev uses seat + credits + overage billing. Zoom uses a bundled-free + premium-feature surcharge model.
The segments with the highest revenue elasticity look strongest today in: enterprise Agent platforms, customer-service Agents, coding Agents, Agent security / identity governance, the iPaaS / MCP / connector layer, and enterprise search / knowledge Agents. They either directly replace labor / outsourcing costs, or lift existing SaaS ARPU and expansion rates, or become the mandatory control point every Agent must pass through before execution.
The companies with the best margins are not necessarily the "sexiest" Agent companies. Better gross-margin / margin prospects usually belong to SaaS platforms that already have distribution channels and ready-made workflows, identity and security control layers, and high-value vertical-industry Agents. The most exposed to inference-cost margin erosion are usually general Agents that are heavily real-time interactive, low priced, high frequency in calls, and lacking workflow pricing power. Microsoft's CFO has stated plainly that the business model for AI applications will increasingly take the form of usage / consumption, while the company hedges inference cost through its hardware stack and software-efficiency optimization.
The segments most prone to valuation bubbles are the Agent-native hot narratives: coding Agents, the general enterprise agent layer, and high-valuation private model companies that lack clear revenue disclosure. Cursor's parent Anysphere reached a valuation near 29.3 billion USD after its November 2025 round. Glean, in its early-2026 official disclosure, had ARR above 200 million USD at a 7.2 billion USD valuation. Harvey's 2026 round reached an 11 billion USD valuation. The valuation expectations for OpenAI and Anthropic run far above traditional software multiples. In public markets, Palantir's current trailing PE already exceeds 150x.
The companies most clearly disrupted by Agents are not all SaaS, but "weak-workflow, weak-permission, weak-data-moat, weak-distribution" seat-based tool software, along with low-value outsourcing, traditional script-based RPA, basic customer-service seats, and parts of low-end development outsourcing. Conversely, companies that own the system of record have a chance to upgrade from a "record system" to an "execution system."
The most important catalysts over the next 12–24 months are not "another Agent launch," but five hard metrics: separately disclosed AI / Agent ARR, acceleration in RPO / cRPO, ARPU uplift from usage-based pricing, customer conversion rates from pilot to production, and the impact of security / governance incidents on purchasing decisions. Gartner has warned that by 2028, 25% of enterprise generative-AI applications will experience at least 5 minor security incidents per year. The Infosys-HFS 2026 survey shows only 14% of enterprises have truly scaled Agentic AI into production.
The biggest risk is not that the technology is "completely unusable," but that enterprise adoption lags narrative expansion: governance, identity, data sovereignty, prompt injection, tool misuse, auditing, and ROI proof will more often become the procurement bottleneck than model capability itself. McKinsey's 2026 AI Trust Maturity Survey likewise notes that enterprises still have clear gaps in governance, risk management, and trust maturity.
Value-Chain Landscape and Technical Architecture
From an investment standpoint, the AI Agent value chain splits into a "capability supply layer" and an "execution control layer." The former includes models, APIs, SDKs, and open-source frameworks; the latter includes connectors, permissions, data retrieval, memory, workflow, governance, security, monitoring, and the final business application entry point. The former more easily faces intensifying competition; the latter more easily forms sustainable customer lock-in and a profit barrier.
| Value-Chain Position | Subsegment | Core Products and Demand Drivers | Revenue Model | Primary Customers | Moat and Margin Characteristics | Representative Companies | Benefit Intensity | Investment Elasticity | Basis |
|---|---|---|---|---|---|---|---|---|---|
| Base models | reasoning, tool use, computer use, multimodal | Support complex planning, tool calls, code execution, browser and desktop operation | token, inference, API calls | Developers, cloud vendors, platforms | Fast iteration but strong price competition; long-cycle moat weaker than distribution and data | OpenAI, Anthropic, Google, xAI, Cohere, Mistral | 5 | 5 | |
| Agent API | Responses API, tool calling, managed tools | Turn the LLM into an executable program rather than a pure chat interface | token + tools + usage | Developers, AI-native apps | APIs are easily standardized; pricing power shifts downstream | OpenAI Responses API, Anthropic Messages/API, xAI Tools | 4 | 4 | |
| Agent SDK / orchestration frameworks | LangGraph, LlamaIndex, CrewAI, Retool agentic workflows | Solve planner, router, memory, subagent, handoff | seat, platform, enterprise support | Dev teams, AI application companies | Strong open-source alternatives; monetization relies on cloud service / observability / hosting | LangChain/LangGraph, LlamaIndex, CrewAI, Retool | 3 | 4 | |
| MCP / connectors | Standardized tool and data access | Let Agents access more systems at low cost | platform fee, connector fee, bundle | Enterprise IT, dev teams | Strong network effects; more connectors make displacement harder | MCP, OpenAI Connectors, Anthropic MCP Connector, Cloudflare MCP governance | 5 | 5 | |
| Enterprise system integration / iPaaS | API management, workflow connector, MCP management | Connect Agents to ERP/CRM/ITSM/DB/file systems | per-connection, per-process, platform subscription | Mid-to-large enterprises | Connector networks, governance, and enterprise implementation capability form a high barrier | Workato, Boomi, Zapier, Make, n8n | 5 | 4 | |
| Data retrieval / enterprise search / RAG | permission-aware retrieval, enterprise search | Reduce hallucination and turn enterprise knowledge into executable context | subscription, seat, usage | Knowledge-intensive enterprises | Permission-aware indexing and cross-system knowledge graphs are the core barrier | Glean, Google Gemini Enterprise, Elastic, MongoDB, Snowflake | 4 | 4 | |
| Memory / context engineering | long-term memory, workflow state, semantic cache | Let Agents retain context across multiple turns and tasks | infrastructure subscription, database, value-added features | Developers and enterprise platform teams | Moderate technical barrier; the commercial moat depends on whether it is embedded in the main workflow | LlamaIndex, MongoDB Vector Search, Snowflake Cortex | 3 | 3 | |
| Identity / permissions / governance | IAM, policy engine, approval, audit trail | The procurement threshold for enterprise Agent adoption | seat, platform, security bundle | Large enterprises, regulated industries | One of the control layers most likely to form an independent profit pool | Okta, ServiceNow AI Control Tower, Workday, Cloudflare | 5 | 4 | |
| Agent security | prompt injection, data leakage, runtime protection | Risk rises sharply once execution becomes autonomous | security subscription, token/API, bundle | Security teams, platform teams | Security is a production must-have, but it needs channel and platform integration capability | Palo Alto Prisma AIRS, CrowdStrike Charlotte AgentWorks, SentinelOne Purple AI | 5 | 4 | |
| Observability / evals | tracing, evals, cost/latency monitoring | Multi-step Agent failures are hard to diagnose manually | seat, usage, enterprise support | AI platform teams | Real demand, but easily eroded by open source; single-point tools are hard to monetize | LangSmith, Arize Phoenix, Langfuse, W&B Weave | 3 | 3 | |
| Enterprise Agent platform | build, deploy, govern, optimize | Unify models, data, tools, permissions, and workflow | seat + credits + platform bundle | Enterprise IT, business units | The most central profit pool; distribution, workflow, and data decide the winner | Microsoft Copilot Studio, Salesforce Agentforce, ServiceNow, Google Gemini Enterprise Agent Platform | 5 | 5 | |
| Enterprise workflow Agent | CRM, ITSM, HR, Finance, Support | Directly replace manual steps and raise process throughput | seat, task, outcome, bundle | Sales, customer service, HR, finance, IT | Industry processes and the system entry point set the ARPU ceiling | Salesforce, ServiceNow, Workday, HubSpot, NICE, Twilio, Zoom | 5 | 5 | |
| Vertical-industry Agent | legal, healthcare, finance, research | High deal value, high compliance, strong data loop | SaaS, usage, enterprise contracts | Law firms, hospitals, financial institutions | High data / compliance barriers; tends toward high gross margin | Harvey, Abridge, EvenUp, Anthropic finance agents | 4 | 5 | |
| Personal productivity Agent | collaboration, documents, meetings, email | Raise individual efficiency and light process automation | seat, freemium, add-on | SMB, knowledge workers | Strong distribution but limited per-customer value; easily commoditized | Microsoft 365 Copilot, Zoom AI Companion, Atlassian Rovo, Google Gemini Enterprise | 3 | 3 | |
| Displacement targets | traditional RPA, low-end service seats, low-value outsourcing | Fixed rules, high repetition, labor intensive | seat or person-day | BPO, IT outsourcers, traditional SaaS | Those lacking data, permissions, and an action entry point are most at risk | Traditional RPA point tools, BPO, low-end dev outsourcing, weak-workflow seat tools | 4 | 5 |
In enterprise Agent architecture, what deserves buy-side analysts' attention most is not "whose model is best," but the following layers. First, the permission and identity layer, because once an Agent starts to "execute," permissions become the product. Second, the workflow and system-of-record entry point, because this decides whether the loop can close. Third, the connector and data layer, because this decides whether the Agent is truly embedded in the enterprise's real systems. Fourth, governance / security / observability, because this decides whether a pilot can reach production. By contrast, pure framework-layer and pure observability-layer products are very good, but without a control plane they easily fall into "great product, hard to monetize." WhyLabs has announced it is ceasing operations, a cautionary counterexample worth noting.
Business Model and Scenario Analysis
Judging from public products and pricing, the Agent business model has already begun departing from traditional seat-only SaaS. The clearest trend is a "seat + credits + outcome" hybrid model: enterprises first buy a platform seat or main product, then pay for tasks actually completed, autonomous run counts, inference credits, resolutions, recommended leads, developer credits, and so on. This model maps more closely to customer ROI than pure token billing, and has a better chance of lifting ARPU, NRR, and retention.
| Pricing Model | Current Representative Companies | Pros | Cons | Impact on ARPU/NRR | Applicable Scenarios | Basis |
|---|---|---|---|---|---|---|
| Per-seat | Microsoft 365 Copilot, Zoom AI Companion, Rovo Dev | Simple to procure, easy to budget, suited to personal productivity | Weakly tied to real business outcomes; seats constrained by headcount | Stable ARPU uplift, but a limited ceiling | productivity, collaboration, dev tools | |
| Per-credit / consumption | Copilot Studio, HubSpot Credits, Atlassian credits | Tied to usage intensity, closer to value | Larger budget volatility; procurement needs cost monitoring | If usage frequency rises, ARPU/NRR elasticity is greatest | enterprise platform, multi-agent workflows | |
| Per-task / outcome | HubSpot Customer Agent / Prospecting Agent, Intercom Fin | Strongest tie to customer ROI; easy to prove labor-replacement cost | Defining "outcome" is complex and needs precise attribution | Most favorable to expansion and retention, especially in customer service / sales | support, sales, collections | |
| Per-process / automation | UiPath, Appian, Workato, Boomi | Maps to enterprise process-transformation budgets; large contract value | Long sales cycles, complex implementation | More favorable to RPO/cRPO | back-office, cross-system processes | |
| Bundled free / upsell | Oracle AI Agent Studio, parts of Zoom/Atlassian AI features | Defends against rivals; fastest to drive main-product retention | Weak direct revenue recognition; easily overstated | More about retention and attach than standalone new revenue | ERP/HCM/collaboration platform defense |
On financial transmission, the Agent impact on SaaS typically comes in three stages. The first stage is "defensive bundling," showing up mainly in win rate and retention rather than new revenue. The second stage is "hybrid monetization," where usage/credits begin to drive ARPU. The third stage is when "system of action" takes hold, meaning enterprises migrate part of their manual-process budget from seats, outsourcing, and person-days to the Agent platform. Salesforce, HubSpot, and NICE have entered the second stage; Microsoft has clearly entered the second stage in personal and developer productivity scenarios; Oracle, SAP, and Workday lean more toward the transition from the first stage to the second.
Inference cost will still compress gross margin, but leading platforms are already hedging along three lines. First, finer-grained routing, pushing simple tasks down to cheaper models. Second, reducing context overhead—using caching, tool search, programmatic tool calling, permission-aware retrieval, and context management to cut token waste. Third, control-plane products (governance, identity, security, workflow) capturing more of total contract value. OpenAI's pricing page already distinguishes input, cached-input, and output costs. Anthropic has launched a Tool Search Tool and Programmatic Tool Calling to reduce context consumption. Google offers simulation, scoring, and optimization in its Agent Platform. Microsoft management has stated clearly that the usage-based business model plus hardware / software efficiency gains will help sustain better AI gross margin over the long term.
Below is a 24-month scenario analysis grounded in current public evidence. It is not an industry fact but a research framework: a way to identify which companies most need "adoption acceleration" to support their current valuation, and which can benefit even with mild adoption.
| Dimension | Conservative | Base | Aggressive |
|---|---|---|---|
| Core assumption | Production still constrained by governance; Agents stay mostly at copilots + limited automation | Customer service, ITSM, sales, and coding keep expanding; usage/credits penetrate rapidly | Agents become the system of action, model cost drops sharply, and governance frameworks mature |
| Enterprise Agent adoption | 15%–20% of large enterprises in production | 25%–35% of large enterprises in production | 40%+ of large enterprises in production |
| Standalone paid rate | Low | Medium | High |
| Inference cost change | Down 20%–30% per year | Down 35%–50% per year | Down 50%+ per year |
| Software ARPU change | Low-single-digit uplift | Mid-to-high single digit to low double digit | Double digit and above |
| Seat count change | Roughly stable | Mild decline in low-end roles/seats | Repetitive seats clearly contract |
| Retention / NRR change | Slight improvement | NRR clearly improves | Retention and expansion improve together |
| Primary beneficiary segments | Platform bundling, governance, security, search | Enterprise Agent platforms, CX, coding, iPaaS, security | Platform layer and workflow-in-control vendors |
| Primary beneficiary companies | Microsoft, Oracle, SAP, Okta, Palo Alto | Salesforce, ServiceNow, NICE, HubSpot, UiPath, Twilio, Atlassian | Microsoft, Salesforce, ServiceNow, Palantir, HubSpot, NICE, Workato/Boomi |
| Primary pressured targets | Pure-narrative Agent startups, weakly differentiated seat tools | Traditional service seats, scripted RPA, low-value outsourcing | BPO, call centers, low-end dev outsourcing, weak-workflow SaaS |
Across these three scenarios, the key watershed is not model capability but two variables. First, whether enterprises are willing to connect Agents to systems that are "writable." Second, whether vendors can move pricing from "AI features given away free" to "charging by completed business outcome." Only when the second variable holds will Agents truly change the SaaS business model.
Segment Breakdown and Competitive Landscape
Below, the focus is on the segments with the most investment value today, rather than every Agent direction that is conceptually plausible. Scores are capped at 5 and weigh commercialization evidence, moat, revenue elasticity, and competitive intensity together.
| Segment | Segment Logic | How Revenue Converts | Current Stage | Pricing Model | Gross-Margin Trend | Moat Core | Future Catalysts | Primary Risks | Investment Appeal | Basis |
|---|---|---|---|---|---|---|---|---|---|---|
| Enterprise Agent platform | Unify models, data, tools, permissions, and governance into one control plane | Expand seats, sell credits, raise attach rate, strengthen RPO | Already in paid-expansion phase | seat + credits + bundle | Mid-to-high, depends on usage mix | Distribution, workflow, permissions, data entry | Standalone AI ARR disclosure, customer production cases | Squeezed by the cloud / model layer | 5 | |
| CRM / sales Agent | Directly connect pipeline, lead, quote, and service handoff | Higher ARPU, higher win rate, more upsell | Among the clearest commercializations | per-seat, per-lead, per-task | High | CRM data + sales workflow | Customer expansion, conversion-uplift disclosure | Hallucination causing business risk | 5 | |
| Customer service / contact-center Agent | Replace basic service seats and outsourcing cost | Outcome billing, AI ARR, seat replacement | Already in large-scale deployment | per-resolution, per-conversation, per-AI-ARR | Higher than outsourcing, lower than pure software platforms | Historical conversation data, process, voice and routing | Resolution rate, human-deflection improvement | Customer-experience volatility, low-price competition | 5 | |
| ITSM / operations Agent | Tickets, changes, approvals, and knowledge bases naturally suit Agents | Lift Pro Plus / premium attach, drive cRPO | Production-leading | high-priced suite + usage | High | Ticketing system, CMDB, approval chain | Large-account count, AI ACV, cRPO acceleration | Large model vendors pushing up into the application layer | 5 | |
| Coding / DevSecOps Agent | Direct ROI, high usage frequency, fast developer diffusion | Lift seats, credits, platform migration | Breakout phase | seat + usage | Mid-to-high, but clear compute pressure | Code repos, CI/CD, dev-process context | Enterprise standardized procurement, rising governance requirements | Open-source substitution, valuation bubble | 5 | |
| Connectors / MCP / iPaaS | To act, an Agent must connect to systems | Platform subscription, process volume, connector upsell | Moving from infrastructure to control plane | platform + flow volume | Very high | Connector ecosystem, enterprise certification, security and audit | MCP standard adoption, multi-system enterprise access | Open standards weakening single-platform lock-in | 5 | |
| Governance / identity / security | A production prerequisite, not optional | Security-budget migration, platform upsell | Heating up fast | security subscription, bundle, API | High | Identity control, policy, audit, runtime protection | Security incidents driving procurement | Possibly absorbed by built-in features of large platforms | 5 | |
| Observability / evals | Multi-step Agents need trace, eval, cost monitor | Engineering budget, enterprise support | Real demand but divergent monetization | seat + usage | Medium | Trace UX, data retention, OpenTelemetry compatibility | Enterprises moving from PoC to production | Strong open-source substitution, price pressure | 3 | |
| Memory / enterprise search / RAG | Decides whether the answer is executable and auditable | Search subscription, platform attach, usage | Entering platformization | seat + platform | High | permission-aware index, knowledge graph | From "found it" to "got it done" | Built into platforms | 4 | |
| RPA 2.0 / Process Orchestration | From script automation to agentic automation | ARR, process volume, platform upgrade | In transition | platform + automation | Mid-to-high | Process orchestration, deterministic execution, governance | Upgrading legacy RPA customers to Agents | Absorbed by platform-type SaaS | 4 | |
| Legal / healthcare / finance vertical Agent | High per-customer value, high labor-replacement cost, strict compliance | High-priced subscription, enterprise contracts | Early to acceleration phase | subscription + usage | High | Professional data and compliance knowledge | More industry cases, paid expansion | Audit liability and regulation | 4 | |
| BPO automation | Directly disrupts the person-day billing model | Replace seats and labor hours | Early, but a clear direction | outcome, seat reduction | High for software side, low for outsourcers | Operations data, process-transformation capability | Large deals in service, finance, procurement | Enterprises reluctant on full automation | 4 | |
| Personal productivity Agent | Massive distribution, low-friction deployment | Seat upgrade, add-on | Mature but medium growth elasticity | seat, add-on | High | Distribution and product integration | Free-to-paid conversion | Commoditization, price war | 3 |
On the competitive landscape, Microsoft, Salesforce, and ServiceNow are currently the three strongest in the enterprise Agent platform layer. Microsoft has the desktop entry, the collaboration entry, and the developer entry, and has already moved Copilot from seat to credits. Salesforce is the first large SaaS company to articulate Agent ARR clearly in public financial terms. ServiceNow's edge is its cross-department process and governance control plane. Google is no weaker in tech stack and Agent Platform completeness—even strong in IAM agent identity, simulation, and governed connectivity—but its control over the office / CRM / ITSM main entry points still trails Microsoft, Salesforce, and ServiceNow. OpenAI and Anthropic have extremely strong influence over models, APIs, MCP, and developer standards, but without deeper workflow and permission entry points, their long-term profit pool more likely stays at "capability supplier" rather than "enterprise operating system." Oracle, SAP, and Workday derive their win rate from owning the system of record: they may not be the first to see visible incremental revenue, but once customers connect Agents directly into ERP/HCM/finance record systems, their stickiness strengthens markedly.
Public-Company Watchlist and Investment Tiering
The table below prioritizes coverage of public companies "most likely to show trackable financial validation over the next 12–24 months." The classification rules: Tier A are core direct beneficiaries; Tier B are clear beneficiaries with valuation, model-cost, or competitive risk; Tier C lean defensive; Tier D have narrative stronger than financial validation; Tier E are potential disruption targets.
| Company | Market | Tier | Core Agent Products | Agent Benefit Path | Observed Commercialization Evidence | Valuation Observation | Research Conclusion | Basis |
|---|---|---|---|---|---|---|---|---|
| Microsoft | US | A | Microsoft 365 Copilot, Copilot Studio, GitHub Copilot | Lift M365 ARPU, drive credits consumption, strengthen Azure and the developer ecosystem | M365 Copilot paid seats exceed 20 million; ARPU growth again led by Copilot and E5; Copilot Studio already billed in Credits; FY25 revenue 281.7 billion USD; current PE about 25x | Not cheap, but still a platform core | Strong beneficiary, among the highest certainty | |
| Salesforce | US | A | Agentforce, Data Cloud, Slack | Directly sell Agent ARR, lift cRPO/RPO, expand existing accounts | FY26 revenue 41.5 billion USD; RPO 72.4 billion USD; Agentforce ARR 800 million USD, +169% YoY; 29,000 cumulative Agentforce deals; current PE about 23x | Versus most AI-narrative stocks, the valuation is still explainable | Strong beneficiary, clearest public financial validation | |
| ServiceNow | US | A | Now Assist, AI Agents, AI Control Tower, Action Fabric | Upgrade large accounts, high-priced suites, governance control plane | Q1 2026 subscription revenue 3.67 billion USD, +22% YoY; cRPO 12.64 billion USD, +22.5% YoY; in Pro Plus AI, the count of customers with ACV over 1 million USD grew 130% YoY | Clear high-valuation platform-stock profile | Strong beneficiary, extremely strong workflow moat | |
| Oracle | US | B | Oracle AI Agent Studio, Fusion AI Agents | Benefit mainly through ERP/HCM retention, module expansion, and cloud usage | Oracle offers AI Agent Studio free to Fusion Applications customers; very favorable for retention, but standalone Agent revenue undisclosed; current PE about 35x | Valuation already clearly re-rated | Clear beneficiary, but leans defensive + cloud/ERP-tied | |
| SAP | US/Europe | B | Joule, Business AI, Autonomous Suite | Use ERP data/permissions to upgrade into a system of action | Q1 2026 current cloud backlog 21.9 billion EUR, +20% YoY; cloud revenue +19% YoY; Joule already live on 35 solutions, but AI revenue not broken out | More balanced than high-valuation US AI stocks | High-quality beneficiary, but incremental-revenue validation still insufficient | |
| Workday | US | C | Illuminate Agents, Agent System of Record, Sana/Workday agents | Defend the HCM/Finance main-system position, lift attach and retention | FY26 Q4 subscription revenue 2.36 billion USD, +15.7% YoY; launched Agent System of Record; the federal HR PAR Agent claims to shorten the PAR cycle by up to 60% | Current PE about 19x, not aggressive for higher-growth SaaS | Strongly defensive; incremental growth elasticity still needs validation | |
| Atlassian | US | B | Rovo, Rovo Dev, AI credits | Deepen cloud-portfolio usage, expand developer/knowledge-worker ARPU | Rovo customer ARR grows about 2x the rate of non-Rovo customers; AI credit usage up 20%+ month over month; Rovo Dev priced at 20 USD/developer/month, overage 0.01 USD/credit | Not yet GAAP-profitable; AI expansion is valuable but volatile | Mid-to-high elasticity; the R&D and knowledge-workflow entry is scarce | |
| HubSpot | US | A | Breeze Customer Agent, Prospecting Agent, Data Agent | Upgrade from traditional seat to outcome billing, fit for SMB/mid-market customers | Q1 2026 customer count 299,500, +16% YoY; average subscription revenue per customer 11,722 USD, +6% YoY; Customer Agent 50 credits per resolution, Prospecting Agent 100 credits per recommended lead | Current PE over 100x; good financial quality but no longer cheap | Strong beneficiary, but the valuation bar is high | |
| Palantir | US | A | AIP, AIP Bootcamps, vertical agentic workflows | Amplify contract value and ARPU via high-value industry/government task automation | Q1 2026 revenue 1.633 billion USD, +85% YoY; US commercial revenue 595 million USD, +133% YoY; full-year revenue guidance 7.65–7.66 billion USD; current PE about 151x | Clearly high valuation, expectations already extreme | High certainty, high elasticity, overheated valuation | |
| UiPath | US | B | Agentic Automation, Robots + Agents + Humans | RPA 2.0 upgrade, existing-customer expansion, oriented to complex process orchestration | FY26 Q4 revenue 481 million USD, +14% YoY; ARR 1.853 billion USD, +11% YoY; DBNRR 107%; gross margin 85% | Current PE about 24x; the market still doubts its long-term competitive position | Clear benefit logic, but heavy competitive and substitution pressure | |
| Appian | US | B | Appian Agents, AI process automation | Use process orchestration to embed Agents into regulated business processes | Q1 2026 cloud subscription revenue 124.5 million USD, +25% YoY; operating cash flow 48.8 million USD | Limited current profitability, small cap, higher elasticity | High-elasticity mid-cap, suited to ongoing tracking | |
| Pegasystems | US | B | Pega Blueprint, workflow + case management agents | Enter agentic workflow via the rules engine and process transformation | Q1 2026 Pega Cloud revenue +30% YoY, Pega Cloud ACV +29% YoY | Current PE about 17x, relatively unaggressive | One of the process-type beneficiaries underestimated by the market | |
| NICE | US/Israel | A | CXone Mpower, AI CX agents | Sell AI directly into contact-center budgets and lift margins | Q1 2026 AI ARR +66% YoY; the company simultaneously raised full-year EPS guidance | Valuation not expensive, but exposed to industry cyclicality | Very strong AI-monetization evidence, worth focused research | |
| Twilio | US | B | AI Agents for support/sales, VoiceAI, TaskRouter | Upgrade communications infrastructure into an AI action layer | Q1 2026 revenue 1.41 billion USD, +20% YoY; operating profit improved markedly; official materials already cite AI Agents for customer support and sales | Current PE very high; the market still treats it more as infrastructure than as a platform | Can transform into a beneficiary, but watch margins and platformization | |
| Zoom | US | B | AI Companion, Custom AI Companion, Virtual Agent | Lift UCaaS/CCaaS ARPU via bundling + add-on | FY26 revenue 4.869 billion USD, +4.4% YoY; 10 of 10 top ZCX deals include paid AI; Custom AI Companion already separately priced | Current PE about 19x; slow growth but AI monetization beginning to show results | Moderate beneficiary, a defense-to-growth watch name |
Scoring Model and Priority
The total scores below are a researcher-model score using user-given weights: Agent revenue direct exposure 25%, product/data/workflow moat 20%, customer quality and revenue certainty 15%, platform ecosystem and connector capability 15%, financial quality and margin 10%, valuation reasonableness 10%, future catalysts 5%. This is not a market-price forecast but a research-priority ranking.
| Company | Agent Revenue Exposure | Moat | Customer Quality | Ecosystem/Connectors | Financial Quality | Valuation Reasonableness | Catalysts | Total | Research-Priority Judgment |
|---|---|---|---|---|---|---|---|---|---|
| Salesforce | 24 | 17 | 14 | 14 | 9 | 7 | 5 | 90 | The core platform name most worth continuing to track |
| Microsoft | 22 | 19 | 15 | 15 | 10 | 6 | 5 | 92 | Highest quality, but both valuation and consensus are high |
| ServiceNow | 22 | 18 | 14 | 13 | 9 | 6 | 5 | 87 | A core enterprise workflow-Agent name |
| NICE | 21 | 16 | 13 | 11 | 8 | 8 | 4 | 81 | Strongest evidence among CX Agents |
| HubSpot | 21 | 14 | 12 | 11 | 8 | 5 | 4 | 75 | SMB/mid-market outcome pricing most worth tracking |
| Palantir | 23 | 18 | 13 | 9 | 8 | 2 | 5 | 78 | Extremely high earnings elasticity, but extremely high valuation risk |
| Atlassian | 17 | 14 | 11 | 12 | 7 | 6 | 4 | 71 | Rovo has room to surprise |
| UiPath | 16 | 13 | 11 | 10 | 7 | 7 | 4 | 68 | Large elasticity if the transition succeeds |
| Pegasystems | 15 | 15 | 11 | 9 | 8 | 8 | 3 | 69 | Low attention, relatively good value for money |
| Appian | 14 | 13 | 10 | 8 | 6 | 8 | 4 | 63 | Small-cap high elasticity, suited to advanced research |
| SAP | 12 | 18 | 14 | 11 | 9 | 7 | 3 | 74 | High-quality defensive beneficiary |
| Oracle | 10 | 16 | 13 | 10 | 9 | 6 | 3 | 67 | Strong retention, incremental yet to be validated |
| Workday | 10 | 17 | 13 | 9 | 8 | 7 | 3 | 67 | More defensive than breakout |
| Twilio | 12 | 11 | 11 | 12 | 7 | 6 | 4 | 63 | In transition, worth watching |
| Zoom | 9 | 10 | 10 | 9 | 8 | 8 | 3 | 57 | Cheap valuation, but Agents are not the core thread |
Global Watchlist
Beyond the focus list above, the following companies are suggested for a secondary research pool. They are either important indirect beneficiaries or geographically notable potential platform-type assets.
| Company/Region | Current Judgment | Brief | Basis |
|---|---|---|---|
| GitLab / US | B | The Duo Agent Platform has taken shape, but revenue disclosure remains insufficient; better tracked via enterprise standardized-procurement progress | |
| Datadog / US | C | AI-driven observability demand is growing, Q1 2026 revenue +32% YoY, but its main benefit comes from infrastructure monitoring rather than the Agent-platform profit pool | |
| MongoDB / US | C | The fusion of vector search and operational data is an Agent-data-layer benefit, but it leans toward infrastructure | |
| Snowflake / US | C | Cortex Agent, AI budgets, and agent-friendly docs show its push in the AI data plane, but the degree of platformization still needs tracking | |
| Okta / US | B | "Okta for AI Agents" cuts straight into the agent-identity control plane—a solid governance-layer beneficiary | |
| Palo Alto / US | B | Prisma AIRS 3.0 directly defines an Agent Security Platform; if enterprise security budgets shift toward Agents, it benefits significantly | |
| CrowdStrike / US | B | Charlotte AI AgentWorks follows a secure-agents ecosystem path, but financial attribution is still early | |
| Cloudflare / US | B | MCP governance, AI Gateway, and Access combine strongly, but currently it is more a platform-infrastructure beneficiary | |
| Alibaba / HK/China ADR | B | Cloud external revenue +40% YoY, AI-related products account for 30% of external cloud revenue—one of the clearest China cloud + Agent commercialization cases | |
| Baidu / HK/China ADR | D | Pushing the Agent narrative and DAA metrics hard, but the public revenue contribution still needs validation | |
| Kingdee / HK | D | The ERP + AI management-platform direction is right, but public quantified financial benefit is insufficient | |
| TCS / India | B | FY26 Q4 annualized AI revenue exceeds 2.3 billion USD, one of the clearest AI-monetization cases in IT services | |
| Infosys / India | C | Topaz/AI Fabric is complete, but it looks more like a service-provider and integrator beneficiary than a software-platform profit pool | |
| HCLTech / India | C | Strong agentic-AI service supply, but the benefit path leans toward project-based services | |
| Capgemini / Europe | B | In 2025 Q4 generative/agentic AI accounted for over 10% of quarterly bookings, with a 600 million EUR intelligent-operations mega-deal | |
| Fujitsu / Japan | C | Clear investment in agentic software development and a secure agent gateway, but financial benefit undisclosed | |
| NTT Data / Japan | C | Advancing vertical-industry Agents jointly with Microsoft/AWS, leaning toward implementation benefit | |
| Samsung SDS / Korea | C | A clear AI Agent strategy and a strong enterprise-work-transformation narrative, but limited global financial validation | |
| NAVER / Korea | D | The "deploy AI agents across all platforms" direction is aggressive, but profitability and commercialization terms are unclear |
Private Companies and Disrupted Industries
Important Private Companies
The table below keeps only the non-public companies "most worth continuing to track." The standard is not fame but whether they hold potential platform power, whether real ARR / paid validation has appeared, and whether they can capture budget.
| Company | Region | Subsegment | Core Products | Funding/Valuation | Revenue/ARR | Relationship to Public Companies | Focus Points | Primary Risks | Basis |
|---|---|---|---|---|---|---|---|---|---|
| OpenAI | US | Base model + Agent API | Responses API, Agents SDK, MCP/Connectors | Extremely high private valuation, needs continued validation | Undisclosed | Deep partnership with Microsoft, also possible to push up into the application layer | Extremely strong standard-setting power | Valuation, governance, and partner dynamics | |
| Anthropic | US | Base model + MCP + coding | Claude, MCP, advanced tool use, finance agents | Extremely high private valuation, to be viewed cautiously | Undisclosed | Ecosystem expansion with AWS, Google, PwC, and others | Strong enterprise coding / governance | Overly high valuation, intense competition | |
| Glean | US | Enterprise search / knowledge Agent | Enterprise search + agentic engine | Series F, valuation 7.2 billion USD | Officially disclosed ARR over 200 million USD | Competing for the entry point with Microsoft, Google, Atlassian, and others | permission-aware enterprise context | Replicated by large platforms | |
| Anysphere / Cursor | US | AI coding agent | Cursor | Valuation about 29.3 billion USD after the 2025 round | ARR not officially disclosed | Competing with GitHub Copilot, GitLab, JetBrains | One of the strongest coding-Agent commercialization segments | Valuation and open-source substitution | |
| Harvey | US | Legal Agent | Legal research, contracts, due diligence, compliance | 2026 valuation 11 billion USD | Undisclosed | Competing/collaborating with Thomson Reuters, RELX, Ironclad, and others | High-value professional-service replacement | Legal liability and compliance | |
| Decagon | US | Customer service Agent | AI customer service agents | 2026 Series D 250 million USD; valuation tripled | Undisclosed | Competing with NICE, Five9, Intercom, Ada | User budget comes from customer service and BPO | Outcome quality and cost | |
| Sierra | US | Enterprise customer service Agent | Enterprise-customized customer service agents | Valuation and ARR widely reported, but still need further validation | Needs further validation | Backed by Bret Taylor, disrupting traditional CCaaS | Strong entry with high-end brand customers | Insufficient public financials | |
| Abridge | US | Healthcare workflow Agent | Clinical documentation, revenue cycle, physician documentation | 2025 round valuation 5.3 billion USD | Undisclosed | Collaborating with the EHR and healthcare-SaaS ecosystem | Healthcare is a high-ROI / high-compliance scenario | Regulation and hospital procurement cycles | |
| Writer | US | Enterprise AI platform | enterprise AI apps/agents | 2024 valuation 1.9 billion USD | Undisclosed | Coopetition with Microsoft, Google, Salesforce | Enterprise workflow platformization potential | Differentiation needs continued validation | |
| LangChain | US | Agent engineering | LangChain, LangGraph, LangSmith | 2025 Series B valuation 1.25 billion USD | ARR undisclosed | Provides the foundation for many application layers | Large standard influence | Balancing open source and commercialization | |
| Workato | US | iPaaS / orchestration | Workato Genie, Enterprise MCP | Valuation needs further validation | Undisclosed | Coopetition with Boomi, Zapier, UiPath, and SaaS platforms | Enterprise orchestration control plane | Low private-valuation transparency | |
| n8n | Germany | Open-source workflow/AI automation | AI agents and workflows | Undisclosed | Undisclosed | Challenges Zapier/Make on open-source low cost | Strong open-source substitution | Monetization ceiling and sales capability |
Traditional Software and Services Disrupted by Agents
What Agents disrupt first is not all software, but products and services whose value proposition has always rested on "clicking, copying, transcribing, answering repetitive questions, and doing weak analysis on a person's behalf inside an interface." The budget-migration path is roughly: low-end service seats / BPO → AI support agent; fixed-rule RPA → agentic orchestration; low-end development outsourcing → coding agents + internal platform teams; weak-workflow seat tools → system-of-action platforms.
| Disrupted Target | Disruption Logic | Which Companies Are at Higher Risk | Which Companies Can Save Themselves | Research Judgment |
|---|---|---|---|---|
| Traditional script-based RPA | Agents can handle semi-structured input, exceptions, and multi-step tool calls | Those that only sell bot seats and lack a workflow and governance layer | Those with orchestration, governance, and robot+agent coordination | UiPath, Appian, and Pega are among the few "beneficiaries amid disruption" |
| Traditional call centers / service seats | AI service begins billing by outcome and entering production | Pure human seats and CCaaS lacking an AI layer | Platforms with routing, knowledge, voice, and an AI layer | NICE/Five9 can upgrade; pure outsourcers face pressure |
| BPO / low-end operations outsourcing | Agents directly replace repetitive person-days | Vendors billing mainly by FTE with many low-complexity processes | Vendors able to transform into AI implementation + managed service | The profit pool shifts from labor to platform + governance |
| Low-end software development outsourcing | Coding Agents reduce person-day demand for simple development/testing/fixing | Pure body-shopping, low-value integration | Service providers that grasp industry processes and platform governance | Indian IT service providers benefit from implementation near term, but face labor-productivity disruption long term |
| Weak-workflow collaboration SaaS | The front-end entry is pulled away by Agents, degrading into a record layer | Tool SaaS lacking a system entry / approval / permissions / data moat | Those with core records, approval flows, and task-execution capability | Asana, monday.com, and similar must guard against the "pushed to the back end" risk |
This means that the system of record will not necessarily be replaced, but many front-end UIs will be rewritten by Agents. If a SaaS vendor owns the core records, permissions, and approval flow, it can upgrade from a back-end record system into an execution system; if it has only a thin interface and a weak data moat, it is more likely to be absorbed by a platform-type Agent. Workday launched its Agent System of Record precisely because it is trying to keep this trend in its own hands.
Risks, Open Questions, and Final Conclusion
The biggest uncertainty for enterprise Agents does not lie in "whether the model is smart enough," but in "whether enterprises are willing to hand it real permissions and real processes." This brings four core risks. First, adoption below expectations: Infosys-HFS data shows only 14% of enterprises have truly deployed agentic AI at scale, meaning the market is still far from broad production. Second, security and governance incidents: Gartner has predicted that by 2028, 25% of enterprise GenAI applications will face at least 5 minor security incidents per year; the fact that Palo Alto, Okta, Cloudflare, and CrowdStrike are all treating Agent security as a new control point shows precisely that this problem will not vanish on its own. Third, model cost and supply changes: if enterprises keep directing budget toward infrastructure and models rather than the application layer, many software companies will face the awkward position of AI capability with "usage but no profit." Fourth, open source and standards: the spread of MCP and open-source observability/frameworks will raise customer optionality and erode the expected premium of certain closed platforms.
Final Judgment
The most important significance of AI Agents within the AI value chain is not creating another "smarter chat box," but pushing enterprise software from the system-of-record era toward the system-of-action era. The long-term profit pool will therefore most likely land in five places: enterprise Agent platforms, the workflow control plane, connectors/MCP/iPaaS, identity and governance, and security and runtime protection. The model layer matters, of course, but it looks more like the upstream compute and capability layer; the application layer can only make real money on the premise of controlling the enterprise's real work.
The five segments most worth attention, I would converge to: enterprise Agent platforms, customer-service / contact-center Agents, AI coding Agents, Agent identity governance/security, and connectors/MCP/iPaaS. These five either already have a clear billing model, or hold the future control plane, or carry the most obvious seat/budget-migration logic.
The ten public companies most worth further research are: Microsoft, Salesforce, ServiceNow, NICE, HubSpot, Palantir, UiPath, Pegasystems, Appian, SAP. Among them, Microsoft, Salesforce, and ServiceNow hold the platform core; NICE and HubSpot are the application layer with the clearest pricing-model change; Palantir is high elasticity but high valuation; UiPath, Pega, and Appian are the group with the most expectation gap in the "process/automation transformation" direction; SAP represents a high-quality defensive beneficiary underestimated by the market.
The ten private companies most worth tracking are: OpenAI, Anthropic, Glean, Cursor/Anysphere, Harvey, Decagon, Sierra, Abridge, LangChain, Workato. If choosing a single narrower thread, I would prioritize AI customer-service Agents and the Agent governance/identity layer: the former commercializes fastest, and the latter is most likely to become a production must-have and form an independent profit pool.
The five points most easily misread by the market are: First, launching an Agent feature ≠ Agent revenue on the ground; Second, strong model capability ≠ the enterprise platform will win; Third, seats may not disappear immediately, but Agents will first lift ARPU and then compress low-end seats; Fourth, security, identity, and audit are not accessories but a procurement threshold; Fifth, open-source frameworks will compress mid-layer pricing, yet they raise the value of control-plane and system-of-record vendors.
The metrics most worth tracking over the next 6–12 months are not "who held another launch event," but: whether AI/Agent ARR is disclosed separately, whether cRPO/RPO accelerates because of Agents, whether the seat + credits structure significantly lifts ARPU, whether the production-customer count grows, whether security/governance incidents change the procurement path, and whether falling model cost improves gross margin.
Suggested narrower follow-on research directions, two as priorities: First, "AI customer-service Agents and contact-center transformation," because the business model has already migrated from human seats toward billing by outcome / by AI ARR, with the clearest financial validation. Second, "Agent governance, identity, and security," because it decides whether enterprises can turn a PoC into a production system, and has the most potential to become an independent new control-layer market.
Open Questions and Limitations
This report prioritizes public materials available as of May 18, 2026, but three categories of limitation must be clearly noted. First, many companies have launched Agent products but have not separately disclosed Agent revenue, ARR, NRR, RPO, or gross margin, so "feature launch" and "financial benefit" must be strictly distinguished. Second, some non-US-listed and private companies disclose only products, funding, or cases, with insufficient revenue and valuation transparency, marked as "undisclosed" or "needs further validation" in the tables. Third, some private model companies and Agent-native companies currently at extremely high valuations have public figures drawn mostly from funding news rather than audited financials, so they are suited as a sector thermometer rather than a basis for high-confidence earnings forecasts.
This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.
Full report
Sign in to read the full report
Sign up free to unlock the full text, the Baillie growth scorecard, and full-text search.
Log in / Sign up free