Market Intelligence Brief Vol. 1 · Compiled Aug 2026 Developer Edition

The AI Agent Report Volume 1

What an AI agent actually is, how many are on sale in 2026, and a field guide to 31 you can buy today — with makers, pricing, autonomy, launch dates, and where to acquire each.

$7.6B
Global market, 2025 (Grand View)
~48%
Consensus CAGR through 2030
67
Agents in the first public census (Dec 2024)
1.3B
Agents Microsoft forecasts by 2028

Figures cross-checked across Grand View, MarketsandMarkets, Emergen, and vendor filings, Jan–Jul 2026. Estimates diverge; see the market section.

01 · Definition

A chatbot answers. An agent acts.

The distinction that matters in 2026 is autonomy. A chatbot responds to one prompt at a time with no memory of a long-term goal. An agent takes a high-level objective, breaks it into sub-tasks, uses tools — a browser, a code interpreter, APIs — and loops until the goal is met.

Ask vanilla ChatGPT “what’s the weather in Paris?” and it answers. Tell an agent “book me a 3-day trip to Paris under $1,200” and it researches flights, compares hotels, and returns an itinerary without further input. The three traits every vendor claims: autonomy (acts without step-by-step guidance), tool use (touches real systems), and goal-orientation (loops toward an outcome).

The autonomy ladder
L1
Assistive
Suggests; human executes. Autocomplete, inline chat.
L2
Supervised
Multi-step with checkpoints and human approval.
L3
Autonomous
Runs minutes to hours unattended, reports back.
L4
Delegated
Brief it and walk away; virtual computer, sub-agents.
Gartner (2026): agents will automate 15–20% of knowledge-work tasks by 2027 — augmenting, not replacing, most roles.
02 · The Market

Small today, forecast enormous — and analysts don’t agree by how much.

Every major research house pegs the 2026 market between $8B and $15B, and every one forecasts a ~45–50% CAGR. Where they part ways is the destination: from $47B by 2030 to $317B by 2035. Read the trajectory, not the decimal points.

Grand View base case · USD billions
$7.6
$10.9
~$47
$183
2025
2026
2030
2033
How the houses disagree
Source2025Forecast
Grand View$7.6B$183B ’33
MarketsandMarkets$7.8B$53B ’30
SNS Insider$7.8B$317B ’35
Roots Analysis$15B$221B ’35
Precedence$7.9B$236B ’34

North America held ~40% of 2025 revenue; Asia-Pacific is the fastest-growing region. Ready-to-deploy agents were ~59% of revenue.

22,000+
Salesforce Agentforce deals
~$1.8B ARR (Agentforce + Data Cloud), Q4 FY26
15M
Paid Microsoft Copilot seats
$5.4B+ ARR from the Copilot/agent layer, Jan 2026
$190M
Harvey ARR (legal)
Nearly doubled in 5 months; ~$11B valuation
03 · The count

“How many specialized agents are for sale?” There is no clean number — here’s the honest range.

No registry lists every commercial agent, and definitions differ wildly — a “product” can be one flagship agent or a platform hosting thousands. The most rigorous public census, the AI Agent Index (the first public database of deployed agentic systems), catalogued just 67 systems as of Dec 31, 2024, filtered to those with real deployment detail.

At the other extreme, marketplace-level counts run into the millions of deployed instances, and Microsoft forecasts 1.3 billion agents by 2028. Between those poles sit the ready-to-buy products a team actually shortlists: realistically several hundred commercial offerings across the categories below. This brief profiles 31 that consistently top 2026 buyer shortlists.

Caveat: counts labelled “30M+ active agents” or “billions” mix deployed instances, API bots, and marketing claims. Treat any single headline number with suspicion — the category has no agreed unit.
04 · The Directory

31 agents you can buy in 2026

Grouped by category. Pricing verified against vendor pages Jan–Jul 2026 — a fast-moving, high-churn space; confirm current rates before buying. “Quote-only” means the vendor publishes no public price.

High autonomy Medium Assistive

General-purpose autonomous agents

brief it, walk away
Agent / Maker What it does Launched Auto. Pricing Limitation / where
ChatGPT Agent
OpenAI
Runs tasks in its own virtual computer with a browser and tools — navigates sites, fills forms, assembles documents. Absorbed Operator (2025). Jul 2025 In ChatGPT Plus $20/mo → Pro $200/mo
Monthly task caps on lower tiers; checks in before consequential actions.
chatgpt.com ›
Manus
Manus (Singapore)
Spins up a dedicated cloud machine with browser, code interpreter and office apps; takes a plain brief and returns a finished deliverable. Topped the GAIA benchmark. Mar 2025 Free (300 daily credits); paid from $20–$39/mo
Individual-focused; credits don’t roll over; Meta acquisition blocked ’26.
manus.im ›
Gemini Agent
Google
Browser automation, deep research and app-connected tasks across Gemini surfaces; absorbed Project Mariner. Strong for Google Workspace users. 2025 Google AI Pro from $20/mo; Gemini Enterprise
Best value inside Google’s ecosystem; Mariner retired as standalone.
gemini.google.com ›
Claude Cowork
Anthropic
General agent that operates on your local file system — reads/writes files, runs multi-step work with safety gating. The most permission-conscious of the three big labs. 2026 Claude Pro $20/mo; Max $100–$200/mo
Most safety-gated computer use; deliberately cautious on actions.
claude.com ›
Genspark Super Agent
Genspark
Mixture-of-agents pipeline purpose-built to produce a website, deck, video or research report from a single prompt. 2025 Free tier; paid subscription
Deliverable-generation focus rather than open-ended computer use.
genspark.ai ›

Coding agents

the fastest-growing role
Agent / Maker What it does Launched Auto. Pricing Limitation / where
Claude Code
Anthropic
Terminal-native agent; multi-hour sessions with sub-agent orchestration and skills-based workflows. Runs Claude Sonnet 5 by default. 2025 Claude Pro $20/mo; Max $100–$200/mo
Weekly usage limits on Pro/Max; token-hungry.
claude.com/product/claude-code ›
Cursor
Anysphere
AI-first IDE with agent-driven multi-file editing (Composer). The developer favorite for fast, in-editor agentic coding. 2023 Pro $20/mo, Pro+ $60, Ultra $200
Daily agent users often hit $60–$100/mo; 2025 pricing backlash.
cursor.com ›
GitHub Copilot
GitHub
Inline completions plus a cloud coding agent that opens PRs. Enterprise standard with IP indemnity and deep IDE integration. 2021 Free; Pro $10/mo; Business $19; Enterprise $39
Moved to usage-based credits Jun 2026; some bills jumped sharply.
github.com/features/copilot ›
OpenAI Codex
OpenAI
One account across CLI, cloud, app and mobile; delegate coding tasks that run autonomously. Runs the GPT-5.6 family. 2025 In ChatGPT Go $8 / Plus $20 / Pro $200
Token-aligned usage billing; cross-surface but newer ecosystem.
openai.com/codex ›
Devin
Cognition
Cloud-hosted autonomous software engineer; parallel agents that plan, code, test and open PRs hands-off. Windsurf folded in as Devin Desktop. Mar 2024 Core from $20/mo; Team ~$500/mo
Highest autonomy, lower throughput per dollar; best for delegation.
devin.ai ›
Antigravity
Google
Agent-first dev platform: agents plan, execute, test and validate across editor, terminal and browser. Model-agnostic (Gemini 3, Claude, OSS). Late 2025 Free platform; pay model usage
New; “mission control” workflow is a learning curve.
antigravity.google ›

Customer-support & sales agents

the biggest revenue slice
Agent / Maker What it does Launched Auto. Pricing Limitation / where
Fin
Intercom
Deflects repetitive support tickets end-to-end; runs standalone on 8+ helpdesks (Zendesk, Salesforce). The transparent-pricing reference point. 2023 $0.99 per resolution; $49/mo base (50 incl.)
Real-world resolution often 42–50% vs 71% vendor claim.
fin.ai ›
Decagon
Decagon
Highly configurable enterprise support agent with strong analytics and sub-second voice; becomes a team’s primary system. $4.5B valuation. 2023 Quote-only; ~$95K–$590K/yr
No public pricing, trial or self-serve; ~6 weeks to deploy.
decagon.ai ›
Sierra
Sierra (Bret Taylor)
Bespoke, deeply branded enterprise conversational agents with white-glove builds; priced per resolved ticket. Serves ~40% of the Fortune 50. 2023 Quote-only; ~$200K–$350K+/yr
Deployments take 3–7 months; enterprise-only economics.
sierra.ai ›
Customer Agent
HubSpot
Suite-native support agent resolving conversations across the HubSpot CRM; aggressive platform-play pricing. 2024 $0.50 per resolved conversation (cut Apr ’26)
Most valuable inside the HubSpot ecosystem.
hubspot.com ›
Ava
Artisan
AI SDR that automates outbound prospecting end-to-end — sourcing leads and sending personalized sequences. Built to replace an SDR seat. 2023 Custom subscription
Outbound-focused; still needs human oversight on messaging.
artisan.co ›
11x (Alice)
11x
Autonomous “digital worker” for outbound sales development — research, personalization and multichannel outreach at scale. 2023 Custom subscription
Data quality and deliverability constraints on cold outreach.
11x.ai ›

Vertical agents

domain-trained, regulation-aware
Agent / Maker What it does Launched Auto. Pricing Limitation / where
Harvey
Harvey
Legal agent for contract review, case-law research, due diligence and drafting; integrates iManage/NetDocuments. Majority of the AmLaw 100. 2022 Quote-only; mid-market ~$1,500–$2,000/seat/mo
Premium pricing debated; dropped its own model for frontier LLMs.
harvey.ai ›
Legora
Legora
Collaborative legal AI for research, review and drafting; hit $100M ARR in 18 months — faster than OpenAI or Cursor at that mark. 2023 Quote-only (enterprise)
No public pricing; enterprise legal focus.
legora.com ›
Abridge
Abridge
Ambient clinical documentation — sits in the exam room and turns conversation into structured notes inside the EHR. $5.3B valuation. 2018 Quote-only (health systems)
HIPAA scope; competes with Microsoft Dragon Copilot’s 200K users.
abridge.com ›
Hippocratic AI
Hippocratic AI
Non-diagnostic healthcare voice agents — pre-op education, chronic-care check-ins, medication reminders. Physician-led safety board; NVIDIA partner. 2023 Quote-only (enterprise)
Deliberately non-diagnostic; ~$3.5B valuation on safety moat.
hippocraticai.com ›
EvenUp
EvenUp
Personal-injury legal agent: automates demand-letter generation and case valuation. ~$2B valuation reshaping a lucrative niche. 2019 Quote-only (law firms)
Narrow vertical; PI-law specific workflows.
evenuplaw.com ›
Vanta
Vanta
Compliance automation agent — continuous evidence collection and monitoring for SOC 2, ISO and audits. ~$300M ARR, the top compliance-AI number. 2018 Quote-only; from low-thousands/yr
Compliance scope; complements rather than replaces auditors.
vanta.com ›

Agent-building & enterprise platforms

build or orchestrate your own
Agent / Maker What it does Launched Auto. Pricing Limitation / where
Agentforce
Salesforce
Ready-to-deploy agents for service, sales and field work, plus a builder over your CRM data. The commercial pace-setter: 22,000+ deals. Oct 2024 $2.00 per conversation; impl. $50K–$150K
Billed per conversation whether resolved or not; Data Cloud dependency.
salesforce.com/agentforce ›
Copilot Studio
Microsoft
Low-code builder for custom agents plus Agent 365 as a control plane for native and third-party agents across Microsoft 365. 2023 M365 Copilot $30/user/mo; consumption add-ons
Most value locked to the Microsoft ecosystem.
microsoft.com/microsoft-copilot ›
Glean
Glean
Enterprise work assistant and agent builder grounded in your company’s permissioned knowledge; SOC 2 Type II. Governance-first. 2019 Quote-only (per-seat, enterprise)
Year-one enterprise deployments can exceed $250K.
glean.com ›
Lindy
Lindy
No-code platform to build business agents that trigger on events and act across your apps (email, CRM, calendar). 2023 From $49.99/mo
Reliability depends on how tightly you scope each workflow.
lindy.ai ›
n8n
n8n
Open-source, self-hostable workflow automation — the most flexible foundation for custom vertical agents with 400+ integrations. 2019 Self-host free; cloud from ~$20/mo
Technical setup; you own the model costs and maintenance.
n8n.io ›
Gumloop
Gumloop
Visual builder for business automation agents with unlimited seats — a fast path from workflow diagram to running agent. 2023 From $37/mo, unlimited seats
Best for structured workflows rather than open-ended autonomy.
gumloop.com ›
Relay.app
Relay
Human-in-the-loop workflow automation with AI steps and approval checkpoints; 100+ integrations across multiple models. 2022 Free AI credits; paid plans
Checkpoint model trades some autonomy for control.
relay.app ›
OpenHands
All Hands AI
MIT-licensed, Docker-sandboxed, model-agnostic open-source agent — runs real autonomous work while keeping code on your hardware. 2024 Free (self-host); pay model tokens
Self-managed; no vendor SLA or support.
all-hands.dev ›
05 · What to watch

Four shifts every buyer is navigating

PRICING

From flat seats to metered outcomes

The market is converging on usage- and outcome-based billing: $0.99 per resolution, $2 per conversation, token-metered credits. Coding tools that flipped to usage-based pricing in 2026 saw some bills jump ~25×. Model your real volume before signing — sticker prices mislead.

AUTONOMY

Trust is falling as capability rises

Confidence in fully autonomous agents dropped from 43% (2024) to 22% (2025); 60% of organizations don’t fully trust them. The winning pattern is human-in-the-loop: the agent drafts, a person approves. Real autonomy still fails on judgment and ambiguity.

VERTICAL

Specialized agents are eating horizontal SaaS

In regulated fields, domain-trained agents beat general copilots: attorney-client privilege, HIPAA, SEC audit trails can’t be bolted onto a raw API. Gartner projects 40% of enterprise apps will embed vertical agents by end of 2026.

BUY vs BUILD

Most buyers partner rather than build

62% of organizations prefer to buy from providers (Salesforce, ServiceNow, SAP); 33% build in-house. A serious mid-market deployment runs $30K–$150K in year one — roughly a third license, a third integration, a third change management.

Sources · consulted Aug 2026

Method: findings synthesized from market-research houses (Grand View, MarketsandMarkets, Emergen, SNS, Roots Analysis, Precedence), vendor pricing pages and earnings filings, and independent 2026 comparison reviews. Pricing and model names shift monthly; every figure is dated where it matters. Where sources conflict — e.g. Fin’s resolution rate or total market size — the disagreement is reported rather than averaged away.