AI Agent Tool Calls Are Distributed Transactions: Who Owns the Retry?

Abivarma R S — Mon, 02 Nov 2026 08:00:16 +0000

An agent with write tools is coordinating distributed transactions over an unreliable network, and nothing limits when it decides to try again. Without end-to-end idempotency, a retry doesn't make it resilient. It just corrupts data faster than any human could.

The failure needs no jailbreak and no hallucination. A refund tool commits, the 200 OK is lost on the way back, the model sees a 504 and calls again. The API sees a fresh request and refunds a second time.

A timeout means "unknown"

Kleppmann makes the point in Designing Data-Intensive Applications: when a request gets no reply, the sender can't tell whether the request was lost, the server died mid-work, or the work ran and the reply was lost. In a naive agent loop the raw error goes straight back into the prompt and the model guesses. Telling it to be careful in the system prompt doesn't make anything idempotent. MCP won't do it for you either: idempotentHint is only a hint, and the spec has no idempotency-key field.

The mechanism: the gateway owns the key

Every state-changing tool call carries an idempotency key that the orchestrator creates, not the LLM. Build it from durable workflow state ({run_id}:{step_id}) so it survives a crash and resume, store a hash of the arguments next to it, and reject a key that comes back with different arguments. A retry with a known key gets the saved response instead of running again, which is how Stripe's API treats repeated keys on the receiving side.

Provider tool-call IDs don't do this job. They match results to calls in the conversation, and they aren't guaranteed to stay the same across regenerations, provider switches or replays.

What I'd act on

Move retry decisions out of the prompt. One layer per dependency owns retries, usually the gateway. If three layers each retry three times, one logical call becomes 27 attempts. Give the model the final outcome, or an explicit "outcome unknown, reconciling" state instead of a raw 504.

Prefer deep tools where the state is local. One transfer_funds call inside a single ACID transaction beats three tools the model has to call in the right order. When a step spans services, use keys plus a transactional outbox, and remember the relay can deliver twice, so consumers still deduplicate.

Don't mistake exactly-once for correct. Keys expire (Stripe may delete them once they're 24 hours old), legacy ERPs often take no key header, and a compensation is a new action, not a rollback. A key makes the wrong refund happen exactly once. It doesn't replace validation, authorization limits or a person reviewing high-value actions.

Let the model propose. Let the gateway commit, once, with a key.

The full article has the diagrams, the gateway code, the sources and the complete checklist: https://abivarma.medium.com/the-dropped-phone-line-why-ai-agent-tool-calls-are-distributed-transactions-7aa85d600d26

Learning AI from First Principles: A Comprehensive Guide

Abivarma R S — Sat, 21 Feb 2026 18:01:13 +0000

Introduction

Artificial Intelligence (AI) is no longer the domain of science fiction; it's a reality that's transforming industries and reshaping how we interact with technology. From self-driving cars to personalized recommendations on streaming platforms, AI's capabilities are expanding rapidly. For developers and tech enthusiasts, understanding AI isn't just a career booster—it's essential for staying relevant in a tech-driven world.

But how do you learn AI from the ground up, especially if you're starting from scratch? This question was recently posed on Hacker News, sparking a vibrant discussion among experts and novices alike. Learning AI from first principles means grasping its foundational concepts before diving into more complex applications. It's about building a solid base that can support your future explorations as AI continues to evolve.

This matters because AI is not a monolithic field; it's a blend of mathematics, computer science, and domain-specific knowledge. Without a strong grasp of the basics, you're likely to get lost in the jargon and miss out on understanding how AI systems truly work. In this blog post, we'll explore how you can start learning AI from first principles, ensuring that you develop a deep and comprehensive understanding of the subject.

Background & Context

The concept of AI has been around since the mid-20th century, but recent technological advances have propelled it into the spotlight. The availability of massive datasets, improved algorithms, and powerful computing resources have all contributed to AI's rapid development. However, the field's complexity often deters newcomers who are unsure where to begin.

The first principles approach, popularized by thinkers like Elon Musk, involves deconstructing complex problems into their most fundamental truths. This method is especially useful in AI, where understanding the basics can provide insights into more advanced topics.

Historically, AI has been divided into various subfields such as machine learning, natural language processing, and computer vision. Each of these areas has its own set of principles and techniques. By focusing on first principles, learners can gain a holistic understanding of AI, enabling them to navigate its many facets with confidence.

Technical Deep Dive

Understanding AI from first principles involves breaking down its core components. Let's explore some fundamental concepts and how they interconnect to create intelligent systems.

Basic Concepts

Algorithms and Data Structures: At its core, AI is about creating algorithms that can make decisions based on data. Understanding basic data structures (like arrays, linked lists, and trees) and algorithms (like sorting and searching) is crucial.
Probability and Statistics: AI systems often rely on probabilistic models to make predictions or decisions. Concepts like Bayes' theorem, probability distributions, and statistical inference form the backbone of many AI algorithms.
Linear Algebra: This branch of mathematics is essential for understanding many AI models, especially those in machine learning. Concepts like vectors, matrices, and eigenvalues are foundational.
Calculus: Understanding derivatives and integrals is crucial, especially when dealing with optimization problems in machine learning.

Machine Learning Basics

Machine learning is a subset of AI focused on creating systems that can learn from data. Here’s a simple pseudocode example to illustrate supervised learning, one of the most common machine learning paradigms:

# Pseudocode for a simple supervised learning algorithm
Initialize model parameters (e.g., weights)
For each epoch:
    For each data point (x, y) in the dataset:
        Make a prediction y_pred using the model
        Calculate the error (loss) between y and y_pred
        Update the model parameters to minimize the loss

In this example, the model learns by iteratively adjusting its parameters to reduce the difference between its predictions and actual outcomes.

Neural Networks

Neural networks are inspired by the human brain and form the basis of many advanced AI systems. They consist of layers of interconnected nodes (neurons) that process input data to produce an output.

Here's a simplified structure:

Input Layer: Receives the raw data.
Hidden Layers: Perform computations and transformations.
Output Layer: Produces the final result.

The power of neural networks lies in their ability to learn complex patterns through training.

Real-World Applications & Use Cases

AI's impact is evident across various industries, each leveraging its capabilities to solve specific problems:

Healthcare: AI systems analyze medical data to assist in diagnosis, predict patient outcomes, and personalize treatment plans.
Finance: AI algorithms detect fraudulent transactions, automate trading, and assess credit risks.
Transportation: Autonomous vehicles rely on AI to interpret sensory data and make driving decisions.
Retail: AI enhances customer experiences through personalized recommendations and inventory management.
Entertainment: Streaming services use AI to recommend content based on viewing patterns.

These applications demonstrate AI's versatility and potential to transform industries by improving efficiency and accuracy.

What This Means for Developers

For developers, understanding AI from first principles offers numerous benefits. It not only enhances problem-solving skills but also opens up new career opportunities. Here’s why developers should care:

Enhanced Problem-Solving: Grasping fundamental concepts allows developers to approach problems more creatively and effectively.
Career Growth: AI skills are in high demand, making developers with these capabilities attractive to employers.
Innovation: A solid understanding of AI empowers developers to create innovative solutions and contribute to cutting-edge projects.

Developers who invest time in learning AI from first principles are better equipped to adapt to the field's rapid advancements and contribute meaningfully to its growth.

Future Outlook

The future of AI is both exciting and uncertain. As technology continues to evolve, so too will the capabilities of AI systems. Here are some trends to watch:

Ethical AI: As AI becomes more integrated into society, ethical considerations will become increasingly important.
Explainable AI: There will be a growing demand for AI systems that can explain their decisions, improving transparency and trust.
AI in Edge Computing: The rise of edge computing will enable AI to process data closer to its source, reducing latency and improving efficiency.
AI and Quantum Computing: The combination of AI and quantum computing could unlock new possibilities in problem-solving and data analysis.

Staying informed about these trends will help developers and tech enthusiasts prepare for the future of AI.

Key Takeaways

Understanding AI from first principles involves grasping foundational concepts in algorithms, probability, linear algebra, and calculus.
Machine learning, especially supervised learning, is a core component of AI that involves training models on data.
AI has diverse applications across industries, transforming healthcare, finance, transportation, retail, and entertainment.
Developers who learn AI from first principles can enhance their problem-solving skills and career prospects.
The future of AI includes trends like ethical AI, explainable AI, and the integration of AI with edge and quantum computing.

Conclusion

Embarking on the journey to learn AI from first principles is a rewarding endeavor. By focusing on foundational concepts, you equip yourself with the knowledge needed to navigate the complexities of AI confidently. Whether you're a developer looking to enhance your skills or a tech enthusiast eager to understand the technology shaping our world, learning AI from first principles is a crucial step.

Ready to dive deeper into AI? Start by exploring online courses, participating in coding challenges, and joining AI communities to connect with like-minded learners. The world of AI is vast and full of potential—your journey is just beginning.

DEV Community: Abivarma R S