The AI agents revolution is creating a massive disconnect between what companies are building and what will actually survive the next few years. At a recent AI Agents Conference in NYC, most vendors were solving problems that may not exist once the technology matures, while missing the real moats that matter in an AI-native world.
Who is it for?
This analysis is essential for startup founders, VCs, and enterprise decision-makers navigating the AI agents landscape. If you're evaluating AI infrastructure investments, building agent-powered products, or trying to understand where sustainable value lies in the AI stack, these insights could save you from betting on the wrong horse.
✅ Key Insights
- ARR per engineer is becoming the critical metric for AI-native companies
- Most current AI infrastructure will be commoditized within 2-3 years
- Trust and risk management emerge as the only durable moats
- Open-source prompt architectures will outcompete closed systems
- Data access standards are rapidly maturing and democratizing
❌ Market Risks
- Massive overinvestment in middleware solutions with short lifespans
- Domain expertise moats are weaker than they appear
- Observability tools face commoditization pressure
- Most "AI-native" companies are building on shaky foundations
- Pricing models based on token markup compress margins quickly
Key Features
The AI agents ecosystem is fragmenting into three distinct layers. The bottom layer consists of commodity LLMs and standardized data access protocols like MCP (Model Context Protocol). The middle layer includes prompt architecture vendors, observability tools, and governance solutions - this is where most conference exhibitors were competing. The top layer focuses on trust, compliance, and risk management, which may be the only sustainable differentiation. Companies succeeding in this space are shifting from engineering-heavy teams to outcome-focused models, with some achieving dramatically higher ARR per engineer ratios.
Pricing and Plans
The traditional SaaS pricing model is breaking down as AI agents commoditize engineering labor. Instead of fixed subscription fees, pricing is moving toward "token markup" models where companies charge 2-4x the underlying LLM costs. This creates margin compression as the core intelligence becomes a commodity. Pricing details may change rapidly as the market matures, but the trend toward outcome-based rather than seat-based pricing appears irreversible.
Alternatives
Rather than buying specialized AI agent tools, many companies are discovering they can "vibe-code" similar solutions in days or weeks using modern AI coding assistants. Open-source alternatives are emerging for most middleware functions, while cloud providers are building native agent capabilities into their platforms. The real alternative may be building internal capabilities rather than relying on external vendors, especially as prompt architectures become more portable and domain expertise becomes more accessible.
Best For / Not For
Current AI agent infrastructure works best for companies in regulated industries where trust and compliance matter more than cost efficiency. It's particularly valuable during the current "magic" phase where integration complexity justifies premium pricing. However, it's not suitable for companies expecting long-term vendor lock-in or those building on the assumption that prompt engineering will remain proprietary. Organizations should avoid betting heavily on closed-source domain expertise or middleware solutions without clear exit strategies.
The AI agents market is experiencing a classic infrastructure bubble where most current solutions will be commoditized within 2-3 years. The only durable moats appear to be trust, compliance, and risk management - essentially turning AI companies into regulated insurance providers. Smart money should focus on the commodity substrate layer or the trust/compliance layer, while avoiding the vulnerable middle layer where most vendors are currently competing.