Selling Agentic AI Has a Problem No One's Talking About

Agentic AI: The Future of Automation?

The Promise and Pitfalls of Agentic AI

• Agentic AI has revolutionized the building process, allowing developers to create functional systems in as little as 35 minutes using tools like Claude in VS Code. However, what can be demonstrated quickly is not necessarily suitable for production environments.

• There is a growing belief that Agentic AI represents the future of automation, but this perspective requires careful consideration to avoid misconceptions that could harm professionals in the field.

• Common claims about Agentic AI include natural language building and self-healing workflows. While these concepts are intriguing, they do not universally apply across all scenarios.

Myth-Busting Common Misconceptions

Myth #1: Self-Healing is the Future

• Although self-healing technology may hold potential, it is currently problematic in production due to unpredictable behavior that can lead to disastrous outcomes (e.g., an agent mistakenly terminating employees).

• For technical users who understand their systems well, self-healing might seem acceptable during development; however, clients require predictability and control over automated processes.

Myth #2: Build Once and Run Forever

• The idea that a system can be built once and run indefinitely ignores real-world challenges such as API failures or changes in program legality (e.g., Ampify's sudden removal of scraping capabilities).

• A truly effective system must incorporate monitoring and alerts to prevent silent failures. Thus, the mantra should shift from "build once" to "build once and monitor forever."

Myth #3: Natural Language Replaces Hard Parts

• While natural language interfaces streamline initial builds by reducing friction, they do not eliminate the need for reliability, auditing capabilities, ownership clarity, or observability within applications.

Speed vs. Delivery: The Real Challenge Ahead

• Agentic AI significantly accelerates prototype development—some projects can be completed in mere minutes—but actual delivery remains complex and labor-intensive.

• In today's market landscape where speed-to-market is crucial, leveraging tools like Agentic for rapid prototyping allows businesses to effectively showcase MVPs without getting bogged down by traditional methods.

Understanding the Importance of Fidelity in Product Development

The Reality of Demos and Final Products

• Demos are inexpensive, but true value comes from understanding how to deliver a reliable product to customers, separating hype from reality.

• A product built quickly is not production-ready; it must be safe, reliable, and consistent for mid-market and enterprise clients.

Determinism and Observability

• The focus should be on determinism—ensuring that the product performs consistently without random errors, aided by observability features.

• Tools like Python scripts can help create deterministic workflows within platforms like Agentic.

Security and Ownership Challenges

• Even with deterministic tools, security concerns remain; it's crucial to ensure that products do not expose users to further errors downstream.

• Hardening a product improves behavior but does not address ownership issues; developers must understand their creations fully.

Trust and Client Relationships

• Selling trust is essential; if developers cannot explain how their solutions work, they risk losing credibility with clients.

• Key questions for client engagement include: Can you review the product? Can the client maintain it? Who will fix issues post-launch?

Business Models for Selling Agentic Solutions

• Understanding business models is critical when selling Agentic solutions. Questions about hosting or maintenance indicate a lack of clarity in business strategy.

• Different business models (product sales, SaaS, managed services, hybrid models) define what is offered to clients and simplify operational complexities.

Simplifying Client Engagement through Clarity

• By selecting a clear commercial model first (e.g., product delivery), businesses can streamline operations and enhance customer understanding of offerings.

Understanding Product Delivery Models

Overview of Product Handover

• When delivering a product, it's essential to assess the client's technical capabilities before handing it over. A simple end build can be easier for non-technical users to understand and maintain.

Managed Service vs. Hybrid Model

• In a managed service model, the provider is responsible for all issues that arise, which can lead to stress due to uptime expectations and service level agreements (SLAs).

• The hybrid model combines product delivery with ongoing support through a retainer, allowing for maintenance and updates while generating recurring revenue.

Importance of Client Education

• Educating clients about the operational model is crucial; if they assume a product will never fail without understanding maintenance responsibilities, it could damage your reputation when issues arise.

Packaging Considerations

• Effective packaging defines scope boundaries (what the product can and cannot do), permissions boundaries (access levels), and data privacy considerations like GDPR compliance.

Risk Management in Product Development

• Understanding failure modes is vital; knowing how different types of failures can impact production helps in preparing for potential issues.

• Ownership clarity ensures that responsibilities for fixing and updating products are well-defined, ideally favoring the provider in a hybrid model.

Monitoring and Observability

• Implementing monitoring systems post-delivery is critical; without visibility into system performance, identifying problems becomes challenging.

Final Steps in Selling Agentic Workflows

• If lacking programming skills necessary for maintaining workflows built on platforms like Python or JavaScript, consider having a senior developer review your work to mitigate risks associated with technical gaps.

Lead Generation System Development

Overview of Development Resources

• The speaker discusses utilizing five LLMs (Large Language Models), seven Python scripts, and one long SOP (Standard Operating Procedure) for a lead generation system. They estimate that an AI-native senior developer could review this in about 3 to 5 days.

• Hiring costs vary by region: approximately $550-$650 for a senior developer in South Africa, cheaper in Eastern Europe, and significantly less in India. This cost-effective approach can provide validation for the project.

Importance of Trusted Reviews

• Emphasizes the need for hiring trusted developers rather than random online hires. Referrals from friends or established businesses are recommended to ensure quality reviews.

• Suggests using multiple LLMs to review each other's code, testing various aspects like security and reliability through specific prompts. This method can streamline the process before human review.

Human Review Necessity

• While AI tools are beneficial, human oversight is crucial for deployment readiness. A senior developer's expertise is necessary to validate findings from automated systems.

Comparing Agentic and Nitn Systems

• Discusses building deterministic systems with Agentic while acknowledging Nitn’s existing capabilities. The focus is on speed to MVP (Minimum Viable Product).

• Highlights Nitn's inspectable workflows which facilitate understanding among non-technical team members, making it easier for internal buy-in compared to more complex systems.

Production Considerations

• Changes in production are simpler with Nitn due to its user-friendly interface, allowing modifications without deep coding knowledge.

• Developers have an advantage as they can leverage these tools effectively; however, reliance on external developers may be necessary for those less technical.

Strategic Recommendations

• Encourages using Agentic for rapid business development while also considering how it can integrate with existing systems like Nitn.

• Notes that while reverse engineering from Agentic might not yield perfect results, it can still provide valuable frameworks or shells for further development.

Infrastructure Considerations

• As projects scale up with agentic builds, AWS infrastructure may become necessary depending on client complexity. In contrast, Nitn offers a self-contained solution suitable even on simple VPS setups.

Understanding the Role of AI in Workflow Optimization

The Shift Towards Automation

• The speaker discusses using a keyboard to activate "whisper flow," emphasizing that the agent automates tasks, reducing the need for traditional problem-solving methods.

• There is a notion that with advanced tools, users no longer need to focus on basic functionalities ("the wheel") as these systems can self-generate solutions.

• The importance of certified development is highlighted; the speaker expresses reluctance to recreate systems without reliable backing and trust in their functionality.

Trust and Reliability in AI Systems

• The discussion shifts to neural networks (NN), where users can delegate some trust to developers, relying on consistent performance from established systems.

• Acknowledgment of inherent risks in any system is made; it's crucial for users to understand that while AI can enhance workflows, it isn't a universal solution or "magic bullet."

Conclusion: Navigating Risks with AI Tools

• Users are encouraged to critically assess the capabilities and limitations of tools like Aentic, recognizing they can improve efficiency but also come with potential drawbacks.