The Agentic AI Stack for Enterprises
- olivermorris83
- Feb 18
- 7 min read
While technologists focus on the implementation of agentic AI, successful leaders are thinking bigger—about ecosystem partnerships, vendor selection, and strategic capability building.
I teamed up with change management expert Simon Torrance (ai-risk.co) and we have developed the Agentic AI Stack—a comprehensive framework that maps this emerging ecosystem from a strategic enterprise perspective.
It reveals both the mature and emerging categories in this new industry. Some layers are bustling with innovation, while others are virtually empty, presenting intriguing opportunities.
The stack spans three tiers:
🎯 Engagement: How AI interfaces with users and systems
💪 Capabilities: The core AI functions and controls
🗄️ Data: The foundational systems that enable learning and accountability
Success with Agentic AI is 20% technology, 80% change management.
The framework puts these opportunities in the right context:
AI-human teams that dynamically self-organize
Scalable intelligence with low marginal costs
Enterprise-wide knowledge amplification
Enhanced operational resilience
The organizations that start building these ecosystem partnerships now will be best positioned to develop competence and build competitive advantage
We'll explore the companies in each category in future posts.
THE STACK
Underpinning the technology elements is the key foundational element: a coherent and robust commercial Agentic AI strategy.
This is a hypothesis that aligns with a company’s corporate strategy. It provides an articulation of the ambition and objectives for Agentic AI, proposes where to deploy it, how to win with it, and the operating model needed to deliver it.
By ‘operating model’ we include not just the technical elements that are laid out in the Stack but also all the non-technical skills, organisational structures, metrics, governance and change processes that will be needed to achieve the objectives.
What is becoming clear is that success with Agentic AI is perhaps 20% technology and 80% change management.
Without these foundational, non-technical, elements in place, articulated in the form of a clear, testable strategic hypothesis, the Tech Stack components are of limited value and potentially dangerous.
Note: We don’t have space in this article to explain in detail each layer and their sub-components, this is reserved for clients. But below we provide a description for each layer, some examples of sub-categories that we think are particularly important and list some companies that are interesting innovators.
1. Engagement Tier
Layer 1: Interfaces
The critical first layer where AI services connect with users, whether customers (consumers or businesses), employees (commercial or IT) or non-human systems (other AI Agents or IoT devices). Should enable natural, controlled interaction while ensuring accessibility and security.
Emerging Sub-Category: Marketplaces & Discovery APIs
Marketplaces and Discovery APIs serve as platforms where AI agents can be discovered, evaluated, and integrated across partner companies, between suppliers and clients. These marketplaces are currently rudimentary, but as ai agents become popular in specialised verticals, the marketplaces will mature to meet demand.
Interesting Innovator: Agent.ai
Created by the founder of HubSpot CRM; a marketplace for AI agents, allowing users to discover, connect with, and hire AI agents for various tasks.
Layer 2: Third-Party Agents
The agents themselves – individual and collective - that enable businesses to serve their end-users in new, more efficient and effective ways.
Emerging Sub-Category: Business-to-AI Agent (‘B2A’)
As AI agents proliferate and acquire the ability to make purchases on behalf of customers and suppliers (within approved limits), they will themselves become customers for dedicated agentic services. Today, this is an entirely new market space. Start-up accelerator, Y Combinator, has recently offered investment for entrepreneurs wishing to kick start this category.
Interesting Innovators: A blank slate, none so far!
2. Capabilities Tier
Layer 3: Controls
The safeguarding layer that ensures AI agents operate safely, ethically and legally within policy boundaries. Implements security guardrails, monitors compliance, and maintains audit trails to build trust and prevent misuse and avoid unintended consequences.
Emerging Sub-Category: Verification & Policy Compliance
The outputs of GenAI models, which underlie agents, are probabilistic hence not 100% accurate. However, it is possible to convert policies and legal documents into a set of rules which can be cross-checked against model output. Such 'formal verification' and policy compliance increases confidence in discuss handing consequential tasks over to AI. There are very few players in this field today, which must expand to fully realise the potential of agentic AI.
Interesting Innovator: Amazon Automated Reasoning
This tool allows users to validate AI-generated content against established policies, identifying inaccuracies and ensuring compliance with business rules. By leveraging automated reasoning, it provides mathematical proof of correctness, reducing the risk of AI hallucinations and enhancing trust in AI outputs.
Layer 4: Orchestration
The coordination engine that manages the way AI agents work with each other and with humans. Handles deployment, monitoring, and workflow management to maximize productivity while maintaining reliability and accountability.
Emerging Sub-Category: Fine-Tuning
Adapting pre-trained AI models to specific tasks or domains by re-training them on specialized datasets. Currently this typically means fine-tuning individual models to a company's specialisms. But research has also demonstrated the ability to fine-tune for agentic teams as the individual agents learn to adopt rich specialisms whilst also collaborating.
Interesting Innovator: Predibase
Offers fine-tuning services for small language models, enabling businesses to customize AI models for their specific use cases.
Layer 5: Intelligence
The cognitive layer that provides AI reasoning and language capabilities, whether from external providers like OpenAI or internal models. Must be designed for flexibility and upgradeability to keep pace with rapid technological advancement.
Emerging Sub-Category: Agent Ops
Encompass the monitoring, error reporting, ongoing evaluation and deployment of agents within an enterprise environment. As AI agents gain autonomy and handle more complex tasks, monitoring their service levels becomes a task to be automated and managed at scale. Agent Ops provides a structured approach to oversee the lifecycle of these agents.
Interesting Innovator: LangSmith
Provides tools to monitor, analyse, and optimize the behaviour of AI agents in real-time. It enhances the observability and reliability of agentic AI applications.
Layer 6: Tools
The action layer that enables AI agents to interact with enterprise systems. Provides secure access to create, read, update, and delete data through APIs or standard user interfaces, including critical capabilities like payment processing.
Emerging Sub-Category: Next Gen RPA & Process Mining
Robotic Process Automation (RPA) and Process Mining are existing technologies which agentic AI can leverage. In enterprise-scale businesses RPA is the traditional mechanism for automating repetitive, rules-based tasks. AI agents can extend that automation work by adding their own artificial reasoning skills. Process Mining, on the other hand, involves analysing business processes to identify complex workflows where inefficiencies are suitable for automation. This helps identify where to apply agentic AI.
Interesting Innovator: UiPath.
A mature business automating repetitive digital tasks typically performed by humans. By combining AI computer vision with APIs, UiPath now enables the creation of AI agents that can understand and execute complex business processes across various enterprise systems.
3. Data Tier
Layer 7: Systems of Record
The foundation technology layer that maintains enterprise memory and ensures continuity. Stores interaction histories, tracks decisions, and manages costs, enabling AI agents to learn from experience and contribute to long-term strategy.
Emerging Sub-Category: Agent Workforce Accounting
Current implementations limit how much agency is granted to agentic systems. But as they become more pervasive and start making financial transactions there will be increasing pressure to track both their direct costs and their decisions. We call this 'Agent Workforce Accounting'. We have found only one player exploring this space today, and see a role for other incumbent accounting systems to expand into this field.
Interesting Innovator: Workday's Agent System of Record
Manages and measures AI agents within an organization. It provides a centralized system to oversee AI agents' activities, ensuring they align with business objectives and compliance requirements.
Other Vendors and Innovators
As part of this work we have carefully built up a database of leading vendors and innovators. Other lists, maps and directories we’ve seen are often incomplete, poorly organised or curated and/or include solutions that are not really agentic.
In the diagram below there's a sample of some other vendors that we think are particularly innovative and/or important across other parts of the stack. There are many more.
Some of the largest tech companies play across multiple layers. We are starting to see a battle to establish ‘control points’ in the market. While heavy vendor investment is ultimately good for enterprise customers, it can also be highly confusing as true vendor capabilities are often difficult to validate.
It is certainly not a good idea to put all your eggs into one basket - a multi-vendor approach makes best sense.
To that end, it will be important for enterprise leaders to consider what interfaces, capabilities, and data are strategic and what are commodities. Then focus on in-sourcing the former and outsourcing the latter.
A robust commercial Agentic AI strategy will help inform those decisions.
Key Innovations and Benefits
The aim of this framework and stack – and future iterations of it – is to enable help support several potentially transformational developments.
Effective AI-Human Collaboration
Teams which can self-organize and dynamically allocate work between human and AI members based on their respective strengths.
Scalable Intelligence
Organizations which can multiply their capabilities without proportional increases in headcount or cost.
Knowledge Amplification
Best practices and expertise which can be captured, enhanced, and deployed consistently across the enterprise.
Operational Resilience
A hybrid workforce which can handle demand surges and complexity while maintaining quality and speed.
Supporting the Emerging Agentic AI Ecosystem
The stack serves as a guide for multiple stakeholders:
For Enterprises:
Provides a roadmap for moving beyond basic Generative AI to true exploitation of Agentic AI at scale and, ultimately, workforce transformation
Helps evaluate and position vendor solutions
Identifies capability gaps and priorities
For Vendors:
Clarifies market opportunities beyond current offerings
Guides product development for the next wave of enterprise AI
Enables integrated solution development
For Investors and Entrepreneurs:
Reveals high-value opportunity areas beyond current GenAI applications
Identifies underserved market segments
Provides context for strategic investments
Looking Ahead
The transition to hybrid AI-human workforces represents perhaps the most significant organizational transformation since the first industrial revolution.
While most enterprises are still focused on implementing advanced analytics and basic Generative AI capabilities, the forward-thinking executives we are working with and talking to are already looking at Agentic Teams as a strategic imperative.
They recognize that future competitive advantage will depend on how effectively organizations can combine human and artificial intelligence.
Our stack has been designed to provide a robust architectural foundation to move beyond current AI implementations towards true hybrid workforces where AI agents and humans work together as teammates rather than tools.
Organizations that begin implementing these patterns now will be best positioned to deliver new levels of value to their stakeholders at near-zero marginal cost, fundamentally changing the economics of their industries.
We believe that the future of work is neither purely human nor purely artificial - it's a seamless collaboration between the two.
The time to start preparing for this future is now.
