New analyst projections suggest the AI agent market will reach $100 billion by 2028. Here's what's driving growth, where the opportunities lie, and what it means for builders.
The forecast: Multiple analyst firms now project the AI agent market will exceed $100 billion by 2028, representing a compound annual growth rate (CAGR) of 45% from current levels. The latest report from Grand View Research pegs 2024 market size at $5.4 billion, with acceleration expected as enterprise adoption matures.
Why this matters: These projections shape where venture capital flows, which problems get solved, and which companies get built. Understanding the forecast helps builders position for where the market is heading.
The builder's question: Is this hype or reality? Where specifically are the opportunities, and what segments are most attractive for new entrants?
Let's examine what's actually being measured and predicted.
Analysts typically include:
Notably excluded: General LLM API revenue (counted separately), traditional RPA (robotic process automation), and narrow-scope chatbots.
The 45% CAGR assumes:
| Factor | Assumption |
|---|---|
| Enterprise adoption | 60% of Fortune 500 deploying agents by 2027 |
| Model capabilities | Continued improvement in reasoning and reliability |
| Integration maturity | Standard protocols (MCP, OpenAPI) enabling interoperability |
| Regulatory clarity | Workable frameworks emerging globally |
| Cost trajectory | 50%+ reduction in inference costs annually |
If any assumption fails significantly, the forecast adjusts accordingly. Model capability improvements and cost reductions seem safe. Regulatory clarity is the wildcard.
| Segment | 2024 share | 2028 projected | CAGR |
|---|---|---|---|
| Cloud/API | 72% | 58% | 38% |
| On-premises | 18% | 22% | 52% |
| Hybrid | 10% | 20% | 68% |
The shift toward hybrid deployments reflects enterprise security requirements and the emergence of capable smaller models that can run locally.
| Use case | 2024 revenue | 2028 projected | Share shift |
|---|---|---|---|
| Customer service | $1.6B | $22B | +5% |
| Sales and marketing | $1.1B | $19B | +3% |
| Software development | $0.9B | $18B | +4% |
| Operations and finance | $0.8B | $16B | +2% |
| Research and analysis | $0.5B | $12B | +3% |
| Other | $0.5B | $13B | -17% |
Customer service leads because the ROI case is clearest and implementation is most mature. But software development is growing fastest - developer tooling is a breakout category.
| Segment | 2024 share | 2028 projected |
|---|---|---|
| Enterprise (>1000 employees) | 68% | 55% |
| Mid-market (100-1000) | 22% | 28% |
| SMB (<100) | 10% | 17% |
Enterprise dominates today because agents require integration with existing systems - something large companies have capacity to implement. As tooling matures, mid-market and SMB adoption accelerates.
Labour economics: In markets facing skilled labour shortages, agents provide scalable capability. A single customer service agent platform can handle inquiries that previously required 50+ human agents.
Competitive pressure: As early adopters demonstrate productivity gains, laggards face competitive disadvantage. This creates adoption cascades within industries.
Capability improvements: Each model generation expands what agents can reliably do. GPT-4 enabled function calling; subsequent models improved reliability to production-grade levels.
Venture investment: Over $15 billion invested in AI agent startups in 2024 alone. This capital is creating products that find markets.
Platform effects: Microsoft, Google, and Amazon are embedding agent capabilities into enterprise platforms. When agents come bundled with existing tooling, adoption friction drops.
Open source momentum: Frameworks like LangChain, AutoGen, and CrewAI lower the barrier to building custom agents. More builders means more solutions means more adoption.
The foundation providers capturing platform economics:
| Company | Position | Moat |
|---|---|---|
| OpenAI | Model + platform | Capability leadership, distribution |
| Anthropic | Model + safety | Enterprise trust, reliability focus |
| Model + cloud | Integration with GCP, Workspace | |
| Microsoft | Distribution + model (OpenAI) | Office, Azure, GitHub ecosystem |
| AWS | Infrastructure + Bedrock | Enterprise relationships, multi-model |
These players capture baseline compute and API revenue regardless of which agent platforms win.
Companies building agent orchestration and deployment:
| Category | Leaders | Challengers |
|---|---|---|
| Horizontal platforms | LangChain, Anthropic Claude | Fixie, Dust, Athenic |
| Vertical solutions | Harvey (legal), Abridge (healthcare) | Multiple per vertical |
| Enterprise orchestration | Microsoft Copilot, Salesforce Einstein | ServiceNow, SAP |
The platform layer is where most value will be captured. Horizontal platforms compete on developer experience and ecosystem. Vertical solutions compete on domain expertise and integration depth.
End-user products built on agent capabilities:
Application layer is crowded but large. Winners will be those who solve real workflows, not those with the flashiest AI demos.
Some industries lag in agent adoption despite clear use cases:
| Vertical | Opportunity | Blockers |
|---|---|---|
| Construction | Project coordination, safety compliance | Fragmented tech stack |
| Agriculture | Yield optimisation, supply chain | Connectivity, data quality |
| Manufacturing | Quality control, maintenance prediction | Legacy systems |
| Local government | Citizen services, permit processing | Procurement, security |
Startups that can navigate industry-specific blockers have less competition and stickier customers.
Several capabilities remain unsolved and valuable:
Multi-agent coordination: Enabling agents to collaborate on complex tasks. Current solutions are brittle.
Long-term memory: Agents that remember and learn from past interactions across sessions.
Reliable tool use: Despite improvements, tool calling still fails too often for mission-critical applications.
Human-agent collaboration: Better interfaces for humans to supervise, correct, and train agents.
Supporting infrastructure that agents need:
Evaluation and testing: How do you know if your agent actually works? Testing frameworks are immature.
Monitoring and observability: What is your agent doing in production? Current tools provide limited visibility.
Cost management: Agent workloads can be expensive. Tooling to optimise costs is nascent.
Security: Agent-specific attack vectors (prompt injection, tool abuse) need specialised defences.
Not everyone agrees with the $100B forecast. Counterarguments:
Capability plateau: What if model improvements slow? Agents become much less compelling without continued capability gains.
Integration friction: Enterprise software integration remains painful. Agents that can't connect to existing systems have limited value.
Trust barriers: Will enterprises trust autonomous systems with consequential decisions? Adoption may be slower than projections assume.
Economic headwinds: If recession hits, AI budgets get cut. The forecast assumes continued investment.
Regulatory risk: Aggressive regulation could constrain agent deployment in key verticals.
These are real risks. The base case forecast likely represents the optimistic end of the probability distribution.
The directional trend is clear: AI agents will be a massive market. Whether it's $50B or $150B by 2028 matters less than the certainty that it's growing rapidly.
For builders, the implications:
The market is real. This isn't speculative technology - enterprises are deploying agents today with measurable ROI.
Platform economics apply. Companies that become platforms where others build will capture disproportionate value.
Vertical depth wins. Generic "AI agent for everything" plays will struggle. Deep expertise in specific workflows creates defensibility.
Infrastructure is undervalued. The picks-and-shovels opportunity (evaluation, monitoring, security) has less competition than application layer.
Timing matters. Too early and you're educating the market. Too late and you're competing with well-funded incumbents. The next 18 months represent a window.
The $100B forecast might be wrong in specifics. But the opportunity it describes is real.
Further reading: