Topic Hub

AI systems

Curated insights and technical field notes on AI systems.

Reader guide

Start with the newest decision note, then follow the topic lanes when you need payments, AI systems, or product architecture depth.

Decision notes

Clear recommendation, rejected paths, and proof trail.

Systems maps

Visual artifacts sit near the section they explain.

Source posture

Public claims stay tied to released work or cited material.

AI systems

Context rot can start before the first prompt

My LLM session loaded 276,989 tool-schema tokens before I typed anything - eleven MCP servers, 609 tool definitions, 27.7 percent of a million-token window spent on payload the model might never call. A bigger context window did not fix that, it gave the problem more room. So I built mcp-broker, PgBouncer for MCP, and the always-loaded tool payload dropped by more than 80 percent.

Lead essay

Start here when you want the current Doric argument instead of a chronological archive.

Read the note

01 / Payments

Commerce architecture

Merchant-of-record decisions, checkout ownership, billing operations, settlement, reconciliation, and risk boundaries.

02 / AI tooling

Practical AI systems

Agentic commerce, multi-agent orchestration, model fine-tuning, runtime governance, and production deployment.

03 / Build notes

From idea to system

The design, architecture, and operating decisions behind Doric Stack products as they move from prototype to public release.

Latest Notes

All

Mar 16, 2026 By Navin Agrawal AI systems 3 min read

AI agents get two payment rails, card and stablecoin

Visa and Coinbase are building competing payment rails for machines, not people. Coinbase's x402 puts stablecoins on HTTP's long-reserved 402 status code for sub-cent agent payments. Visa keeps the card rails and their dispute resolution. The split is real, because a 10,000-call-an-hour agent and a $4,000 flight want opposite things.

Feb 16, 2026 By Navin Agrawal AI systems 3 min read

AI can suggest a payment, only the control plane completes it

A payment bot did not go rogue - the architecture did. A timeout produced two valid-looking instructions for the same payout, because a downstream retry never learned the first was already in flight. Mandate-based authorization like AP2 still needs a runtime control plane with real state discipline, not policy in a config file.

Dec 1, 2025 By Navin Agrawal AI systems 2 min read

Agents do the wiring, architects carry the tradeoffs

AI agents are arriving faster than most banking career ladders can adapt, and the market is already rewarding people who work with them. The career moat was never wiring up another API call. It is the judgment calls agents cannot own - which rail to trust, how much latency to spend, who the regulator will be.