Agent Skills
Ross Sylvester, Co-Founder & CEO, Adrata | Feb 2026 | ~14 min read
In November 2025, Gartner made two predictions that sound contradictory until you understand what they actually mean.
Prediction one: by 2028, AI agents will outnumber human sellers 10:1.
Prediction two: fewer than 40% of sellers will report that AI agents improved their productivity.
Both will be true. Here is why, and what it means for every CRO evaluating AI investments right now.
What a Skill Actually Is
The word "agent" has been abused beyond recognition by marketing departments across the industry. Every chatbot, workflow trigger, and automated email template has been relabeled an "agent" in the last twelve months. This is not helpful.
An agent, in the technical sense, is a system that can perceive its environment, reason about what to do, take actions, and learn from outcomes. A chatbot that sends a templated response is not an agent. An agent is something else entirely. It pulls details from a prospect's company. It reads their recent LinkedIn posts. It identifies a relevant business problem, drafts a personalized message referencing that problem, decides the optimal send time based on timezone and past engagement, then adjusts its approach based on whether the prospect opens, clicks, or replies. That is an agent.
A skill is what makes an agent useful in a specific domain. A skill is packaged expertise -- the domain knowledge, behavioral patterns, and operational instructions that transform a general-purpose AI into a specialist. An agent with an "account research" skill doesn't just summarize a company's About page. It synthesizes firmographic data, recent earnings, leadership changes, job postings that signal growth or contraction, technology stack indicators, and competitive landscape -- and it knows which of those signals matter for a CRO versus a marketer versus a product leader.
Skills are how agents avoid the trap of being good at everything and great at nothing.
The distinction matters because the quality of the skill -- the depth of domain expertise embedded in it -- determines whether the agent is genuinely useful or just fast at being mediocre. Speed without accuracy is worse than manual work. A bad email sent instantly is still a bad email.
The Landscape: Who Has What
Every major GTM platform shipped agent capabilities in 2025. Here is what each one actually does, based on what's in production -- not what's on a roadmap slide. A caveat: the performance statistics in this section are vendor-reported. Independent verification is limited. Read them as directional, not gospel.
Salesforce Agentforce
Salesforce is making the largest platform bet. Their Agentforce SDR handles inbound lead qualification -- summarizing web-to-lead submissions, identifying buying intent, suggesting next steps. Their Sales Coach agent supports mid-to-late stage activities: meeting prep, deal hygiene, closing strategy. In December 2025, Salesforce acquired Qualified, adding autonomous pipeline generation via an AI SDR agent that engages website visitors.
The internal numbers: Salesforce's own pilot contacted 130,000 previously untouched leads in four months, generating 3,200 qualified opportunities.
HubSpot Breeze
HubSpot's Breeze platform includes four purpose-built agents. Their Prospecting Agent functions as a 24/7 BDR -- monitoring buying signals, researching accounts, personalizing outreach. Their Closing Agent answers product and pricing questions during the deal cycle without waiting for a rep. HubSpot's Spring 2025 product launch included over 200 updates anchored around AI agents, and their INBOUND 2025 theme was "building hybrid human-AI teams."
A telling detail from HubSpot's customer conversations: some customers have requested to see AI agents listed alongside human employees in performance dashboards. They want to manage both in the same view.
Gong
Gartner named Gong the Leader in the 2025 Magic Quadrant for Revenue Action Orchestration, placing it highest for Ability to Execute. Their agent suite is built on conversational intelligence: an AI Deal Reviewer that suggests deal updates and eliminates manual CRM data entry, an AI Tracker that extracts revenue-specific concepts from customer interactions, and an AI Theme Spotter that identifies recurring patterns across the entire business.
The data: companies using Gong's AI Tracker achieve 35% higher win rates. Gong's CRO Shane Evans shared at Battery Ventures' 2025 event that conversational intelligence tools now give each rep back nearly a full workday per week.
Outreach
Outreach offers three agents targeting critical sales cycle moments. Their Revenue Agent identifies high-intent accounts, sources contacts, and crafts personalized messages. Their Research Agent pulls insights from conversations, meetings, and external sources. Their Deal Agent eliminates manual updates by automatically populating CRM fields based on call signals.
Key metric: sellers using Outreach's AI tools cut research and personalization time by 90%. Their conversation intelligence shaves 11 days off sales cycles and boosts win rates by up to 10 percentage points on deals over $50K.
Salesloft + Clari
The Clari and Salesloft merger closed in August 2025, creating what they call a "Revenue AI powerhouse." The combined platform offers 26 AI agents spanning prospecting, deal management, and pipeline efficiency. Their Fall 2025 launch included a Sales Strategist Agent for personalized coaching and an Influence Graph that automatically maps stakeholders with sentiment and influence scoring.
But Salesloft's own research reveals the gap between aspiration and reality: only 6% of sales leaders believe their teams have the skills to use sales AI effectively. Only 6% of sellers use AI for task prioritization. The technology is ahead of the adoption.
Clay
Clay has positioned itself as infrastructure for GTM agent skills, trusted by over 300,000 teams. Their Claygent research agent has surpassed one billion runs. Unlike traditional enrichment providers that return static database results, Claygent browses websites, reads pages, and returns insights in real time at scale.
Clay also created the "GTM Engineer" role in 2023 -- a hybrid role combining sales operations, data engineering, and AI workflow design. Today, approximately 100 GTM engineer job listings go live every month at companies like Cursor, Lovable, and Webflow.
Apollo.io
Apollo has evolved from a data and sequencing platform to an AI-native sales platform with agentic outbound (AI-generated research, scoring, and outreach automation) and agentic deal execution (conversation capture, pre-meeting summaries, auto-generated follow-ups, and CRM field population). They operate on a copilot-to-autopilot spectrum, letting teams choose their level of AI autonomy.
ZoomInfo
ZoomInfo's Copilot Workspace, launched October 2025, is an AI-powered execution engine handling research, follow-ups, signal monitoring, outreach drafting, CRM updates, and next-best-action surfacing. Their customers report a 14-point increase in close rates -- from 32% to 46%.
6sense and Demandbase
Gartner has named 6sense Leader in ABM for five consecutive years. Their platform unifies data, intelligence, activation, and orchestration through AI agents. Their research reveals that 94% of buying groups rank preferred vendors before first contact, purchasing from that favorite 77% of the time. Demandbase launched Agentbase in March 2025, going beyond account-level intelligence to individual stakeholder engagement within buying committees.
The Skills Taxonomy
Across all these platforms, a clear taxonomy of GTM agent skills is emerging:
Top-of-Funnel Skills:
- Account research -- synthesizing firmographic data, news, filings, job postings, tech stack, and growth signals into actionable briefs
- Prospect enrichment -- real-time enrichment beyond static databases, including web browsing and dynamic context
- Signal detection -- monitoring buying signals across web visits, email engagement, content consumption, product usage, job changes, and third-party intent data
- ICP qualification -- AI-powered filtering and scoring against ideal customer profile criteria
- Outreach personalization -- generating messages that reference specific account context, buyer signals, and pain points at scale
Mid-Funnel Skills:
- Meeting preparation -- compiling intelligence briefings with company news, stakeholder profiles, past engagement, and competitive mentions
- Buyer group mapping -- automatically mapping every stakeholder within an account with sentiment and influence analysis
- Competitive intelligence -- continuous monitoring of competitor activities, providing deal-specific counter-positioning
- Deal scoring and risk detection -- analyzing conversation signals and CRM data to flag risk and predict outcomes
Bottom-of-Funnel Skills:
- Pipeline management -- automatic deal updates, CRM field population, stage progression based on real signals
- Sales coaching -- real-time and post-call coaching, call scoring, and development recommendations
- Forecasting -- predictive models combining engagement data, conversation signals, and historical patterns
- Follow-up generation -- contextual follow-up emails and next-step recommendations from meeting content
The gap in most platforms: buyer group intelligence. The ability to map not just individual contacts but the full buying committee -- who holds authority, who influences, who blocks -- and to track how the group's dynamics evolve across the deal. 6sense's data makes this critical: when 94% of buying groups have ranked their preferred vendor before first contact, understanding the committee composition before your first meeting is not a nice-to-have. It is a prerequisite.
Skills vs. Automation: The Critical Distinction
Salesloft provides the clearest framework for this distinction. Traditional automation executes predefined, deterministic steps -- known rules, fixed paths, templated actions. An agent skill assesses live inputs, adapts to real-time situations, makes decisions, and takes action.
The test: if the system needs manual intervention to handle exceptions, it is automation. If it can assess context and adjust its approach, it is an agent.
The practical difference for a CRO:
Automation: "When a lead fills out the form, send email template A, wait 3 days, send email template B." Same message, same cadence, every time.
Agent skill: "When a lead fills out the form, research their company, identify their likely business challenge based on their role and industry, reference a relevant case study, and adjust the follow-up timing based on their engagement pattern with the first message." Different message, different cadence, for every lead.
The most advanced organizations use agents to orchestrate workflows -- the agent decides which workflow to trigger, with what parameters, based on contextual assessment. This is the convergence point that matters.
What the Data Actually Shows
The research on agent-based selling versus traditional selling is consistent -- and nuanced.
Salesforce surveyed 4,000+ sales professionals for their State of Sales 2026 report. 87% of sales organizations now use some form of AI. 54% of sellers have already used AI agents. 89% plan to by 2027. High performers are 1.7x more likely to use AI agents for prospecting than underperformers. But here is the number that matters most: the average seller still spends only 40% of their time actually selling. Agents are expected to cut prospect research time by 34% and email drafting by 36%.
McKinsey's "Agents for Growth" analysis (2025) found that effective agent deployments could deliver 3-5% annual productivity improvements and 10%+ growth lift. AI-driven personalization enhances customer satisfaction by 15-20%, increases revenue by 5-8%, and reduces cost to serve by up to 30%. Some Fortune 250 brands have accelerated campaign execution by 15x.
BCG (2025) found 35% of organizations already using agentic AI, with another 44% planning to do so soon. Three-quarters of extensive adopters believe AI is enabling new sources of competitive advantage.
Outreach's 2025 Sales Data Report identified lead qualification as the number one seller challenge. 45% of teams already use a hybrid AI-SDR model. Sellers using AI tools cut research and personalization time by 90%.
The broader statistics: companies adopting AI in sales saw productivity jump 30%, revenue grow 13-15%, and sales cycles become 68% shorter. ZoomInfo customers report a 14-point increase in close rates.
But then the Gartner contradiction: despite all these gains, fewer than 40% of sellers will report productivity improvement by 2028. The explanation is not that agents don't work. It is that most organizations deploy agents without redesigning workflows, without cleaning data, and without training teams on how to work with AI. They bolt agents onto broken processes and expect transformation. They get noise.
The OpenAI Case Study
Perhaps the most striking real-world example comes from Battery Ventures' 2025 CRO gathering. OpenAI's revenue team has zero SDRs. Their internal workflow agent, nicknamed "Taylor," qualifies inbound leads, parses intent, and closes smaller deals autonomously.
Ashley Kramer, OpenAI's Head of Revenue, stated: "I probably have 5x the target with one-fifth the team."
But she also stated that multimillion-dollar enterprise deals still require her personal involvement. The agent handles the operational throughput. The human handles the relationship and judgment.
This is the operating model. Not agents replacing sellers. Not sellers ignoring agents. Sellers and agents each doing what they do best -- with the boundary drawn at the point where human judgment is irreplaceable and agent throughput is superior.
What CROs Should Actually Evaluate
Based on the research, here is what matters when evaluating agent platforms:
1. Data foundation. 74% of sales professionals are focusing on data cleansing to maximize AI returns. High performers are 1.5x more likely to prioritize data hygiene than underperformers. If your CRM data is dirty, agents amplify the mess. Start here.
2. Skill breadth. Does the platform cover the full GTM taxonomy -- top, mid, and bottom of funnel? Or is it a point solution with one skill rebranded as a "platform"? The tools that win are the ones that connect signals across the entire revenue cycle.
3. Autonomy spectrum. Can you dial between copilot (human reviews every action) and autopilot (agent operates independently)? Binary systems -- fully autonomous or fully manual -- don't fit real revenue organizations where different tasks require different levels of human oversight.
4. Buyer group intelligence. Can the platform map and engage entire buying committees, not just individual contacts? Given 6sense's finding that 94% of buying groups rank preferred vendors before first contact, contact-level intelligence is no longer sufficient.
5. Explainability. Can the agent explain why it made a recommendation? Black-box scoring without rationale erodes trust and prevents calibration. The rep needs to know why the agent thinks a deal is at risk -- not just that it is.
6. Outcome measurement. Does the platform track agent-attributed pipeline, revenue, and productivity? If you can't measure the agent's contribution, you can't optimize it.
7. The "agent washing" test. Ask the vendor: does this system make autonomous decisions based on contextual data, or does it execute predefined templates? If it needs manual intervention to function, it is automation wearing an agent costume.
The Skill That Matters Most
Gartner surveyed 1,026 B2B sellers in 2024 and found that the highest-impact competency for quota attainment was AI partnership. Sellers who effectively partnered with AI were 3.7x more likely to meet quota. This beat tactical flexibility (3.4x) and ability to read buyers (2.9x).
The top skill in sales is no longer product knowledge, objection handling, or even discovery. It is knowing how to work with AI.
This is a learnable skill. It requires understanding what the agent is good at (data processing, pattern recognition, exhaustive research, consistent follow-through) and what it is not good at (reading rooms, building trust, navigating organizational politics, making judgment calls under genuine ambiguity). The sellers who calibrate this boundary accurately outperform the ones who either over-trust the agent or ignore it.
Salesloft's research shows that only 6% of sales leaders believe their teams are ready. The other 94% need a skills development plan that is as structured and deliberate as any sales methodology rollout.
Agent skills are not a software feature. They are an operating model transformation. The CROs who treat them accordingly will build revenue organizations that compound their advantage. The ones who treat them as another tool purchase will wonder why the productivity gains never materialized.
The agents are coming. The question is not whether your team will work with them. It is whether your team will work with them well.