Best AI Agents for Enterprise in 2026: 14 Agents Mapped Across 7 Functions

The best AI agents for enterprise in 2026 aren't one product you roll out company-wide. They're a portfolio: a customer service agent from Decagon or Sierra, an IT helpdesk assistant from Moveworks, a revenue agent from Gong or Clari, a finance agent grounded in the ERP you already run, each bought separately, deployed in a deliberate order, and judged against its own function rather than a single company-wide scorecard. This guide maps 14 ready-made agents, agents you deploy rather than platforms you build on, across the seven functions where large organizations are actually buying them: service and support, IT and employee helpdesk, sales and revenue, finance, HR, security, and knowledge work. Every price and certification claim was checked against a vendor's own site in August 2026, and anything not vendor-published is labeled "(reported)" with a named source.

That's a different question than the one our best enterprise AI agent platforms guide answers. That guide is written for a CIO or CISO running a platform through procurement and security review, the infrastructure layer you build custom agents on. This one is written for whoever decides which finished agent goes live in which department, and in what order, once a platform question is settled or beside the point entirely. If your team is weighing build versus buy, best AI agent platforms covers that separate decision in full. And if what you actually need is software that assists a person rather than a system that plans and executes several steps on its own, our best AI tools for enterprise guide covers that different, larger category instead: an AI tool drafts something for a person to send or approve, while an agent plans a sequence of actions, calls tools to execute them, and checks in with a human only at defined boundaries.

Updated August 2026. Pricing and certification claims below were checked against vendor sources this month. Agentic products change pricing models and ownership fast enough that we recommend re-confirming anything cost-critical directly with the vendor before you budget against it.

Key Facts

• Enterprises buying ready-made AI solutions instead of building their own rose from 53% in 2024 to 76% in 2025, per Menlo Ventures' State of Generative AI in the Enterprise report.
• Only 11% of organizations are actively running agentic AI in production today, with 30% still exploring and 35% reporting no formal agentic strategy at all, per Deloitte's 2026 State of AI in the Enterprise survey.
• Gartner projects over 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls, not model failure, as the leading causes.
• 62% of executives expect more than 100% ROI from agentic AI investments, averaging 171% expected return, according to PagerDuty's 2025 Agentic AI Survey of 1,000 executives, a projection almost nobody has banked yet.
• 29% of employees admit to actively working against their company's AI strategy, a figure that rises to 44% among Gen Z, per an April 2026 survey of 2,400 knowledge workers by Writer and Workplace Intelligence, reported by Fortune.
• 52% of knowledge workers use unapproved AI tools despite company policy, even though 90% of executives believe they have visibility into what's being used, per an Okta-commissioned Apprize360 survey of 292 executives and 492 knowledge workers.
• Best-of-breed software procurement fell to 20.7% of enterprise buyers in early 2026, down from 24.3% six months earlier, while 41% of organizations are now actively planning to consolidate applications onto fewer platforms, with AI cited as a leading driver, per Futurum Group's 1H 2026 Enterprise Software Decision Maker Survey of 830 IT decision-makers.

Quick Comparison Table

Buy the Agent, Not Just the Platform

Most enterprise AI coverage in 2026 answers a procurement question: which platform certifications, which governance model, which system of record. That's real work, and it's necessary before you sign anything. But it isn't the question a COO asks when the board wants to know what's actually running by next quarter. That question is narrower and more concrete: which finished agent goes into which department first, who owns it once it's live, and how do you know in six months whether the whole program is working, not just one pilot.

The 14 agents below answer that question. None of them are frameworks you configure from scratch. Each is a product with a specific job: resolve a support ticket, triage a phishing alert, draft a forecast, answer an HR policy question. Menlo Ventures' 2025 enterprise AI research found that buying ready-made solutions instead of building them jumped from 53% to 76% of enterprises in a single year, and the function map below is built around that reality: most large organizations are shopping a portfolio of finished agents, not commissioning fourteen custom builds.

Where to Deploy Your First Agent (and Why That Order Matters)

The most common mistake in enterprise agent rollout isn't picking the wrong vendor. It's picking the wrong function to go first. A finance or HR agent often gets funded first because it has the biggest visible budget line, and that's exactly backwards: it's also where a wrong autonomous action costs the most, and where your organization has the least practical experience running an agent through a real rollout cycle. Deploying AI agents to production lays out the staged rollout every agent should pass through, shadow mode, limited human-gated traffic, risk-gated full traffic, then full rollout, and where you set those gates should track the cost of a mistake, not the size of the budget behind the project.

The org that runs IT helpdesk or customer service through a full rollout cycle first isn't being cautious for its own sake. It's earning the governance pattern, the guardrail templates, the monitoring habits, the honest post-mortem, that its finance and HR agents will need later and that a first attempt almost never gets right on the first try.

Function 1: Customer Service and Support

Customer-facing resolution carries the highest stakes of any function on this list, because a bad agent decision is visible to the person paying you, not just to an internal reviewer. The three agents enterprises actually buy here split on how they ground that risk.

ServiceNow AI Agents

ServiceNow grounds customer service agents in the same CMDB and case data its ITSM and CS Management products already hold, with an AI Control Tower for governance across connected systems. Since completing its acquisition of Moveworks on December 15, 2025, ServiceNow has folded Moveworks' front-end assistant into the same platform, though that integration is still settling. Pricing runs through a custom quote; since April 2026, ServiceNow replaced its older tiers with three AI-native ones, Foundation, Advanced, and Prime, with Prime the only tier carrying fully autonomous agents and still rolling out through 2026. Third-party estimates put core tiers around $90 to $200+ per user per month (reported, unconfirmed by ServiceNow directly). ServiceNow holds ISO 27001, ISO 27017, ISO 27018, SOC 2, FedRAMP, and IRAP certifications.

Sierra

Sierra, co-founded by Bret Taylor and Clay Bavor, builds conversational agents purpose-built for customer experience across chat, voice, and email, and nothing else; it isn't built for internal workflows. Its outcome-based pricing, getting paid when the agent actually resolves something rather than per seat or per message, is unusual on this list, and unresolved conversations typically aren't charged. Reported per-resolution rates cluster around $1 to $2.50, with realistic year-one cost, implementation included, often reaching $200,000 to $350,000+ (reported, unconfirmed by Sierra directly). Sierra holds PCI DSS Level 1 Service Provider certification, the highest payments-industry tier, alongside SOC 2 Type II, ISO 27001, and ISO 42001.

Decagon

Decagon positions itself as Sierra's most direct challenger, with a similar usage-based philosophy and no per-seat floor, though there's no self-serve signup for either; every deployment starts with a sales conversation. Reported figures include an annual platform fee around $50,000 plus per-conversation fees around $0.50 to $0.99, with a reported median contract near $400,000 a year and starting deals around $95,000 (reported, unconfirmed by Decagon directly). Decagon holds SOC 2 Type II and is GDPR compliant, with HIPAA available under a signed BAA; ISO 27001 and PCI DSS weren't publicly listed as of this research.

For the fuller, vendor-neutral ranking of this function alone, including outcome-based challengers beyond these two, see best AI agents for customer service.

Function 2: IT and Employee Helpdesk

The employee-facing helpdesk is where most large organizations run their first genuinely agentic deployment, because the job is bounded, the volume is high, and getting a password reset wrong costs a support ticket, not a customer relationship.

Moveworks (a ServiceNow Company)

Moveworks built its reputation on front-end AI assistance for employee IT, HR, and facilities questions through what it calls a Reasoning Engine, and it's proven at real enterprise scale pre-acquisition. Pricing is quote-only, billed per employee per year on annual or multi-year contracts; reported figures range from roughly $50,000 to $100,000 a year at 500 to 1,000 employees, up to $200,000+ above 5,000 employees, with a reported median around $130,000 (reported, aggregated third-party buyer data, unconfirmed by Moveworks directly). Moveworks holds SOC 2 Type 2, ISO 27001, ISO 27017, ISO 27018, ISO 27701, ISO 42001, CSA STAR Level 2, FedRAMP, GDPR, and CCPA; HIPAA wasn't listed on its own security page as of this research.

Microsoft 365 Copilot Agents

Separate from Copilot Studio, the low-code builder, Microsoft ships a set of finished agents directly inside Microsoft 365 Copilot. The Employee Self-Service Agent answers HR and IT policy questions grounded in your own SharePoint and Viva content and is still expanding its rollout. Alongside it, the Researcher and Analyst agents, which reached general availability on June 2, 2025, cover deeper knowledge work: Researcher synthesizes information across email, files, chats, and the web, while Analyst runs on OpenAI's o3-mini reasoning model to work through data questions with Python. All three ship inside the same $30/user/month Copilot seat license, billed yearly, so an organization already paying for Microsoft 365 Copilot gets employee helpdesk and knowledge-work agents without a second procurement cycle. Certifications inherit the Azure and Microsoft 365 compliance umbrella: SOC 2, ISO 27001, HIPAA, and FedRAMP High P-ATO.

See the full IT support ranking for self-serve alternatives below the enterprise tier this guide focuses on, including Atomicwork, Workativ, and Freshservice's Freddy AI Agent.

Function 3: Sales and Revenue

Sales and revenue is the most crowded function on this list, and the one where "which agent" genuinely depends on which slice of the job you're automating: the CRM itself, the calls that feed it, or the forecast that comes out of it.

Salesforce Agentforce

Agentforce reasons directly over live Salesforce data, accounts, cases, opportunities, through Agentforce Builder and its Atlas Reasoning Engine, which removes the integration layer most other agents on this list still need to build. Pricing runs on three separate models: Flex Credits at $500 per 100,000 credits (about $0.10 per action, $0.15 per voice action), a Conversations model at $2 per conversation for customer-facing agents, or per-user licensing from $125 a month; Enterprise Edition orgs get 100,000 complimentary Flex Credits through Salesforce Foundations. Salesforce publishes one of the broadest certification sets on this list: SOC 2, SOC 3, ISO 27001, ISO 27017, ISO 27018, ISO 42001, HIPAA, FedRAMP, PCI DSS, GDPR, and CSA STAR.

Gong AI Agents

Gong grounds its agents in recorded sales conversations rather than CRM fields alone, and its February 25, 2026 Mission Andromeda launch expanded that into Gong Enable for enablement, a conversational Gong Assistant that answers questions grounded in real calls, and an MCP Server that lets outside AI tools, including Salesforce Agentforce, Microsoft Copilot, Claude, ChatGPT, and Gemini, query Gong's deal and call data directly. Gong doesn't publish pricing; its own pricing page confirms a per-user license plus a platform fee, both quoted after you submit team size. Third-party buyer reports put per-user pricing around $120 to $250 a month, plus a $5,000 to $50,000 platform fee and $15,000 to $65,000 in implementation (reported, figures vary meaningfully by source, unconfirmed by Gong directly). Gong holds SOC 2 Type II, ISO 27001, ISO 27017, ISO 27018, ISO 27701, ISO 42001, HIPAA, PCI DSS, and GDPR.

Clari Revenue AI Agents

Clari's Revenue AI Agents sit inside a broader revenue orchestration platform (deal scoring, forecasting, pipeline inspection, sales engagement) and, since an April 2026 MCP Server launch, expose that same live pipeline and call data to outside tools like Claude, ChatGPT, Microsoft Copilot, Gemini, and Agentforce. Clari doesn't publish pricing either; its own pricing page confirms a quote-based model with no extra platform fee for integrations. Reported figures put per-user pricing around $1,000 to $2,000 a year, most mid-market deals near $1,300, plus $15,000 to $75,000 in implementation, and buyer reports note that two companies the same size can pay 40% more or less than each other for a comparable plan (reported, unconfirmed by Clari directly), a pricing-opacity gap worth pressure-testing before you sign. Clari holds SOC 2 Type II, ISO 27001, HIPAA, and GDPR/CCPA compliance.

See best AI agents for sales for the full 15-agent ranking across prospecting, outreach, and qualification, functions this map doesn't cover in depth.

Function 4: Finance

Finance agents earn their keep by grounding in the ERP or financial dataset an organization already trusts, because the alternative, an agent reasoning over a synced copy of financial data, is exactly the kind of integration gap that turns a routine reconciliation into an audit finding.

SAP Joule

Joule is SAP's copilot and agent layer, embedded across S/4HANA Cloud, RISE with SAP, and GROW with SAP, with Joule Studio available for building custom agents on top. Its advantage for an SAP-standardized finance team is grounding directly in ERP objects, orders, invoices, master data, without a separate integration layer. Baseline conversational Joule is generally included with current SAP cloud subscriptions at no separate license fee, metered in AI units, with reported overage around $0.08 to $0.18 per action beyond the included allowance; Joule Studio custom-skill development is reportedly priced around $42,000 to $96,000 a year (reported, unconfirmed by SAP directly). SAP's Trust Center lists SOC 1 and SOC 2 reports and ISO/IEC 42001 certification specific to AI management systems; FedRAMP status wasn't independently confirmed for Joule or RISE in this research.

Workday Illuminate

Illuminate is Workday's agent system spanning both HR and Finance workflows, recruiting, contingent sourcing, document-driven accounting, supplier contracts, grounded in the same HCM and Financials dataset Workday already holds. That dual grounding is why it appears in both this function and HR below: one agent system, two functions, because Workday treats HR and Finance data as one connected model rather than two separate products. Access runs through Flex Credits included in a Workday subscription and renewed annually; base enterprise Workday subscriptions are reported from roughly $34 per employee per month (reported, base platform pricing, not Illuminate-specific). Workday holds SOC 1/SOC 2, a consolidated ISO/IEC 27001 certificate, FedRAMP Authorization at the Moderate impact level for Workday Government Cloud, and a HIPAA third-party attestation.

For accounts payable, receivable, and audit specialists beyond these two ERP-grounded agents, see best AI agents for finance teams.

Function 5: HR and People

HR agents carry a different risk profile than the rest of this list: the questions they answer and the decisions they touch (leave policy, benefits, hiring) sit closer to regulatory and works-council scrutiny than a CRM update ever will.

Workday Illuminate, covered above under Finance, is also the deepest HR-grounded agent on this list: its recruiting, screening, and case-management agents run against the same HCM dataset that powers its Financials agents, so an enterprise already standardized on Workday effectively buys both functions from one vendor relationship. Microsoft 365 Copilot's Employee Self-Service Agent, covered above under IT and Employee Helpdesk, does the same double duty from the Microsoft side: it's built to answer HR policy questions (leave balances, benefits enrollment, expense rules) with the identical grounding and governance model it uses for IT questions, because both are really the same job, an employee asking a policy question and getting a governed, sourced answer, pointed at two different departments' documents.

That overlap is worth naming plainly before you shortlist a third, HR-only vendor: if you already run Workday or Microsoft 365 at enterprise scale, you likely have most of an HR agent capability sitting inside a contract you've already signed, and the honest first move is confirming what it can do today before buying a dedicated HR agent platform on top of it. Best AI agents for HR covers the purpose-built HR helpdesk and cross-functional specialists (Leena AI, Resolve, Rippling AI, Darwinbox, and others) worth comparing against that baseline.

Function 6: Security

Security is the least mature function on this list for a genuinely freestanding, vendor-agnostic agent purchase. Most enterprises aren't buying one standalone "security agent" the way they buy a customer service or sales agent. They're extending the agents already embedded in the security stack they run.

Microsoft Security Copilot

Microsoft's approach ships "embedded agents," prebuilt, role-specific agents built directly into products a security team already uses rather than a separate console: the Security Alert Triage Agent in Microsoft Defender (formerly the Phishing Triage Agent, now extended to identity and cloud alerts), a Conditional Access Optimization Agent in Microsoft Entra, a Vulnerability Remediation Agent in Microsoft Intune, and a Data Security Agent in Microsoft Purview. At St. Luke's University Health Network, the Security Alert Triage Agent cut phishing-alert review time by more than 200 hours a month, triaged alerts up to 78% faster, and caught 6.5 times more malicious email, per Microsoft's own Security Copilot blog, a vendor-published case study rather than an industry average. Pricing runs through standalone Security Compute Units at $4 an hour, pay-as-you-go, or bundled at no extra cost inside Microsoft 365 E5 and E7. Security Copilot inherits Microsoft's Azure compliance umbrella: SOC 2, ISO 27001, and FedRAMP High P-ATO.

That's a narrower list than the other six functions on purpose. Dedicated, vendor-neutral SOC agent products, autonomous Tier-1 triage, threat hunting, and investigation built by security-first vendors rather than embedded in a productivity suite, are a real and fast-moving category of their own, and covering them properly needs more room than one section of a function map can give them. See best AI agents for cybersecurity for that fuller, 13-agent ranking, including CrowdStrike Charlotte AI, Torq, and Google Security Operations.

Function 7: Knowledge and Expert Work

The knowledge function covers two related but distinct jobs: finding the right answer across a fragmented app estate, and reasoning like a domain expert inside one specific, high-stakes profession.

Glean

Glean is an enterprise search and Work AI platform that indexes across every connected app, with Glean Agents built on the same permission-aware index, so an agent's search results respect the same access controls a person's would. That's a real differentiator against ecosystem-native agents (Microsoft's, Google's, Salesforce's), which ground well inside their own suite but less well across everything else a large, acquisitive enterprise typically runs. Glean doesn't publish a price list and reports a roughly 100-seat minimum; third-party reporting cites a base seat fee around $45 to $50 a month plus a separate AI add-on around $15 a month (reported, unconfirmed by Glean directly). Glean confirms SOC 2 Type II, HIPAA, and GDPR directly on its own security page, states ISO/IEC 27001 and ISO/IEC 42001 certification, and maintains zero-retention agreements with its model providers so customer data isn't used for training.

Google Gemini Enterprise

Renamed from Agentspace in October 2025, Gemini Enterprise bundles a chat-style front end with an Agent Development Kit for building custom agents, grounded across Google Workspace plus more than 100 third-party connectors. Google doesn't publish a reachable, current pricing page; reported figures put the Business edition around $21 a seat monthly on an annual commitment and Standard around $30 a seat monthly on a 12-month term, or $35 without one (reported, confirm current numbers with Google Cloud sales before budgeting). Gemini in Workspace apps was the first generative AI assistant for a productivity suite to reach FedRAMP High authorization, and Gemini Enterprise inherits Google Cloud's SOC 1/2/3, ISO 27001/27017/27018, ISO 42001, and HIPAA (with a signed BAA) certifications.

Harvey

Harvey is the clearest example on this list of a domain-expert agent rather than a general search layer: built specifically for legal work, grounded in research through a LexisNexis partnership (Ask LexisNexis), and reasoning over case law, contracts, and filings in ways a general enterprise search tool isn't built to do. It reached an $11 billion valuation in March 2026 and $190 million in annual recurring revenue by January 2026, serving 100,000+ lawyers across 1,300 organizations, including the majority of the AmLaw 100, more than 500 in-house legal teams, and named enterprise clients including PwC, A&O Shearman, HSBC, and NBCUniversal. Its March 2026 Agent Builder lets firms construct their own task-specific agents on top of the platform. Harvey doesn't publish pricing; it's negotiated per seat monthly through enterprise sales, with a roughly 20-seat minimum on an annual contract. Reported figures put mid-market firms (50 to 200 attorneys) around $1,200 to $1,500 a seat monthly, AmLaw 100 firms at $1,500 to $2,000+, and total enterprise contract value between $50,000 and $200,000 a year (reported, unconfirmed by Harvey directly). Harvey holds SOC 2 Type 2, ISO 27001, ISO 27701, and ISO 42001 certification, GDPR, CCPA, and IRAP, and became one of the first AI agent products in legal to earn AIUC-1, a certification built specifically for AI agents rather than adapted from traditional software audits; HIPAA isn't listed among its certifications.

Microsoft's Researcher and Analyst agents, covered under IT and Employee Helpdesk above, also belong in this function for any team already on a Microsoft 365 Copilot seat: the same $30/user/month license that answers an HR policy question also synthesizes research and runs data analysis, one more example of a single vendor relationship quietly covering more than one row on this map.

Certifications at a Glance

Certifications don't guarantee a safe deployment, but a missing one is a fast way to get an RFP disqualified, and it's worth knowing where each of these 14 stands before your security team asks. "Not confirmed" means a claim wasn't independently verifiable in the vendor's own published material, not that the control doesn't exist.

Vendor Consolidation or Best-of-Breed: How to Actually Decide

Futurum Group's 1H 2026 survey of 830 IT decision-makers found best-of-breed procurement has fallen to 20.7% of buyers, down from 24.3% six months earlier, with 41% of organizations now actively planning to consolidate applications onto fewer platforms and AI cited as a leading driver. That trend shows up directly in the function map above: Workday and Microsoft each quietly cover two functions from one contract, and ServiceNow covers customer service and IT helpdesk from the same case-grounded workflow. The practical question isn't whether consolidation is happening; it's whether it's the right call for your specific function.

Neither answer is the "serious" one. An enterprise running Moveworks because its employee IT questions genuinely needed a dedicated assistant is making just as sound a decision as one running Microsoft's Employee Self-Service Agent because it's already paying for the seat. The mistake is picking based on which vendor happens to be easiest to reach, not which one actually fits the function.

Change Management Is the Real Failure Mode

At enterprise scale, the agent that fails from a broken tool call is rare. The agent that fails because employees quietly route around it is common, and it's the failure mode this list of vendors can't fix for you. 29% of employees admit to actively working against their company's AI strategy, rising to 44% among Gen Z, according to an April 2026 survey of 2,400 knowledge workers by Writer and Workplace Intelligence, reported by Fortune; of those who admitted to it, 30% cited fear that AI would take their job as the primary reason. Separately, an Okta-commissioned Apprize360 survey found 52% of knowledge workers use unapproved AI tools despite policy, even as 90% of executives believe they already have visibility into what's being used, a confidence gap that shows up as shadow AI rather than open resistance.

The agentic workforce names the deeper pattern underneath both numbers: treating an agent as a coworker without building the oversight that framing implies creates overtrust (reviewers rubber-stamp what they've learned to trust), skill atrophy (junior staff lose the reps that used to build their expertise), and a blurry accountability line (nobody can say who owns it when the agent gets something wrong). Each shows up in a portfolio of agents faster than in a single pilot, because there are more handoffs, more reviewers, and more places for "that's not really my job anymore" to go unaddressed.

Measuring a Portfolio of Agents, Not Just One

A single pilot can survive on "it seems to be helping." A portfolio of agents across seven functions can't, because at that point a CFO is going to ask what the whole program is worth, not just the one agent someone happened to demo well. AI agent ROI names the gap directly: governance overhead, human review time, audit trail management, incident response, can add 30% to 50% to the apparent savings of a single, well-run agent, and that discipline gets harder, not easier, to maintain once you're running fourteen agents across seven departments instead of one.

That measurement gap is exactly what sits behind PagerDuty's finding that executives expect an average 171% return from agentic AI even though Deloitte's 2026 research found only 11% of organizations are actually running agentic AI in production. Most of that expected number describes a projection, not something anyone has actually banked yet, and a portfolio spanning seven functions is the wrong place to keep running on projection alone.

Enterprise Agent Rollout Mistakes to Avoid

How to Choose: Decision Framework

| If you need... | Start with... | Why |

The safest enterprise shortlist starts with the job, the system that owns its data, and the cost of a wrong action.

|---|---|---| | A safe, low-stakes first agent to prove the model works | An IT helpdesk or Tier-1 support agent | High volume, bounded scope, existing ticket data to shadow-test against | | Agents grounded in a CRM you already run | Salesforce Agentforce | No integration layer between the agent and live CRM data | | Deal and forecast intelligence layered onto sales calls | Gong or Clari | Both ground agents in real conversation and pipeline data, not CRM fields alone | | An employee-facing IT and HR assistant | Moveworks or Microsoft 365 Copilot's Employee Self-Service Agent | Purpose-built for the questions employees actually ask, not a general chatbot | | Agents grounded in ERP or HCM data you already trust | SAP Joule or Workday Illuminate | Native object-level grounding without a new integration project | | SOC-facing triage without buying a separate security platform | Microsoft Security Copilot's embedded agents | Ships inside Defender, Entra, Intune, and Purview rather than a new console | | One search and agent layer across a fragmented app stack | Glean | Permission-aware search across everything connected, not one ecosystem | | A dedicated agent for expert knowledge work in a regulated profession | Harvey, for legal | General search grounding isn't the same as domain-specific reasoning over case law and contracts | | Customer-facing resolution at enterprise scale | Sierra or Decagon | Purpose-built for the payments and compliance bar customer-facing agents actually need |

---

---

What to Do Next

Pick one function, the lowest-stakes, highest-volume queue you have, and run exactly one agent through a full rollout cycle before you shortlist a second. Use the rollout sequencing table above to confirm you're starting in the right place, not the function with the loudest internal sponsor. Once that first agent has cleared shadow mode, limited traffic, and a real measurement window, you'll have the governance pattern, the reporting cadence, and the honest sense of what a portfolio actually costs to run responsibly, all three of which matter more to your second, third, and fourth agent than any feature on this page.