标签：#ai

Most teams test email by not testing it. The send path gets a mock — expect(transport.send).toHaveBeenCalledWith(...) — and everyone agrees that's "good enough." The receive path gets skipped entirely, because there's no honest way to assert on a real inbox from a test runner. So the one part of your system that talks to the outside world over an unreliable, asynchronous, third-party channel is the part with the least coverage. That's backwards. The reason email is hard to test isn't the sending. It's the asserting . You can fire POST /messages/send all day, but to prove the message actually left, rendered correctly, and arrived with the body you expected, you need a real mailbox you control — one you can read programmatically and throw away when the run finishes. Shared Gmail test accounts almost get you there, but they bring OAuth on the runner, catch-all races between parallel workers, and a 90-day token that expires the night before a release. This post is about a different fixture: a disposable Agent Account created at the start of a CI run and deleted at the end. You mint a real mailbox per run (or per test), point your application at it, send and receive real mail, assert on the actual message body, and tear the whole thing down. No OAuth. No shared inbox. No leftover state. What an Agent Account gives you here An Agent Account is just a Nylas grant with a grant_id . That's the whole trick, and it's worth saying plainly because it's what makes this pattern cheap: an Agent Account works with every grant-scoped endpoint you already know — Messages, Drafts, Threads, Folders, Attachments, Webhooks. There's nothing new to learn on the data plane . If you've ever called GET /v3/grants/{grant_id}/messages , you already know how to read a test inbox. The difference from a normal grant is provisioning. A regular grant needs a real human to complete an OAuth flow. An Agent Account is created with a single API call — no OAuth screen, no refresh token, no human. It's a m

Most "AI email agent" demos quietly assume the agent answers everything. Point a model at the inbox, generate a reply, send it, repeat. That's a fine loop right up until the model hits a message it shouldn't touch — an angry customer, a legal question, a refund the agent has no authority to approve — and confidently fires off a reply anyway. The expensive failures in agent email aren't the threads the agent gets wrong. They're the threads the agent answers at all when it should have stepped back. So let's build the part that steps back. Not the classifier that decides a message is risky — that's triage , a separate problem. This is the handoff : once something flags a thread as "needs a human," how do you actually pull the whole conversation out of the agent's reach, park it where a person can find it, and make sure the agent keeps its hands off until that person clears it? I work on the Nylas CLI, so the terminal commands below are the exact ones I reach for when I wire up an escalation path. Every operation gets the two-angle tour: the raw curl call and the nylas command that does the same thing. What the handoff actually needs An Agent Account is, underneath, just a Nylas grant with a grant_id . That's the spine of everything here, and it's worth sitting with: there is nothing new to learn on the data plane. The same grant-scoped endpoints you already use — Messages, Threads, Folders, Drafts — work against this grant exactly the way they work against any Gmail or Microsoft grant you got through OAuth. So the escalation path isn't some special agent feature. It's three plain operations you already half-know: A place to put escalated threads. A custom folder — call it Needs human — that lives alongside the six system folders every Agent Account ships with ( inbox , sent , drafts , trash , junk , archive ). A way to move the whole thread there. Not one message — the thread . A reply is just the latest message in a conversation; a reviewer needs the full chain. A way

Most "AI email agent" demos end with a triumphant send . The model writes a reply, the code POSTs it, and a real message lands in a real stranger's inbox. That's a great demo and a terrible production default. The moment your agent can send mail with nobody watching, you've handed an LLM a corporate email address and the standing authority to use it. One hallucinated price, one confidently wrong refund promise, one apology to the wrong customer, and you're explaining to legal why a bot signed an email as your company. There's a boring, durable fix that predates AI by decades: don't send — draft. Stage the message, put a human in front of it, and only send once someone with a name and a pulse approves. Email systems have had a "Drafts" folder forever for exactly this reason. The Nylas Drafts API turns that folder into something better — an approval queue your agent writes into and your reviewers drain. This post builds that queue. The agent creates a draft, a human reviews the pending drafts, and an approved draft gets sent byte-for-byte unchanged . No re-rendering, no "the agent regenerates it on approval" race where the thing you approved isn't the thing that ships. I work on the Nylas CLI, so the terminal commands below are the exact ones I reach for, and I'll pair every one with the raw curl so you can wire it into a backend in whatever language you like. This is deliberately not about escalating inbound threads to a human (that's a different problem, where the trigger is a message arriving). Here the trigger is the agent wanting to send , and the gate sits on the outbound path. Why a draft is the right approval primitive You could build approval a dozen ways. You could buffer the agent's output in a queue table and call send later. You could stash a JSON blob in Redis. Both work, and both quietly reinvent something the email stack already gives you. A draft is a real, persisted email object , on the mailbox, with a stable id . That buys you three things a homegr

DoorDash RAG Architecture, AI Agent Mesh, & Open-Source Supply-Chain Scanner Today's Highlights This week, we explore advanced AI agent orchestration, a detailed production RAG architecture, and an open-source tool for supply-chain security auditing. These stories provide practical insights into deploying and managing AI frameworks in real-world workflows. How DoorDash Built an AI Shopping Assistant That Doesn’t Rely on the LLM Alone (InfoQ) Source: https://www.infoq.com/news/2026/07/doordash-ai-ask-assistant/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=global This article from InfoQ delves into the intricate architecture behind DoorDash's "Ask DoorDash" AI-powered shopping assistant. Unlike many solutions that solely depend on large language models, DoorDash's approach integrates an LLM with a complex retrieval-augmented generation (RAG) system and a comprehensive intent classification pipeline. This multi-layered framework ensures accuracy and relevance, particularly for tasks like recommending specific items or answering detailed product queries within their extensive catalog. The system also employs sophisticated filtering and ranking mechanisms to refine results, moving beyond simple keyword matching to provide highly personalized and context-aware suggestions. The technical deep-dive covers how DoorDash engineered this system to handle the nuances of user intent and data retrieval efficiently in a production environment. Key aspects include leveraging structured and unstructured data sources, managing latency for real-time interactions, and implementing robust feedback loops for continuous improvement. The article offers valuable insights into building scalable, reliable AI assistants that can augment LLMs with proprietary data and business logic, providing a blueprint for enterprises looking to deploy similar advanced applied AI solutions. Comment: This provides a fantastic real-world case study for augmenting LLMs with custom RAG and

This is a condensed version of my preprint ( DOI: 10.5281/zenodo.21345310 , CC BY 4.0). Reference implementation: askbar.pro . The library problem For thirty years the website has been a library: a visitor arrives with one question and is expected to find the answer themselves, navigating menus, pages, and filters. Visitors read a small fraction of site content. Most leave without doing the thing the site owner hoped for. Chat widgets bolted onto such sites change nothing: the maze remains, the widget just answers questions about the maze. The pattern The Librarian Pattern inverts the relationship. The site does not present itself; it asks what you need and assembles the answer. The bar as the primary interface. One persistent input, text and hold-to-talk voice. It replaces navigation. Scene reassembly (generative UI). The center of the screen is not a page but a scene, composed per recognized intent. Transitions morph rather than reload. A guide with a plan. The conversational layer is a consultant with a goal ladder, asking one next question, never presenting menus of three options. Two button systems. Global suggestion chips above the bar are visually separated from in-scene action cards. This prevents the "six buttons" degeneration of chat UIs. The static shadow. Every live scene has a server-rendered twin page: full text in the DOM, question-shaped headings, FAQ schema, llms.txt, freshness stamps. Humans get the agent; crawlers and AI answer engines get complete, citable pages, generated from the same content source. Structural GEO-readiness. Content already organized as questions and answers matches how generative engines retrieve and cite, by construction. The result that surprised me 24 hours after the discoverability layer went public, Yandex Alice (the largest Russian AI answer engine) began citing the reference implementation as its prime example for the "next-generation website" query, describing the mechanics correctly and distinguishing it from "a chat

Of all the debates raging about the potential downsides of AI, there is one worry causing the most hand-wringing among AI enthusiasts in Silicon Valley — that the giant AI labs that sell proprietary models are somehow acting like Trojan horses.

iOS 27 escaped the developer world today with the launch of the first public beta. I've been testing the new operating system since early June, looking for quirks and seeing if it can live up to the hype Apple promised in the keynote. This year's iOS upgrades are what one might call a Snow Leopard […]

Ukraine’s drone blitz halted Russia’s Sea of Azov shipping in under a week.

VistralNova began by developing Web3 gaming experiences and is now expanding toward developer...

Apple would not comment on the "security breach," which allegedly allowed a former employee to download sensitive files from Apple's network long after he departed the company for rival OpenAI.

OpenAI accused of conspiring with former Apple employees to steal trade secrets.

Startle the snake by striking the grass. — The 36 Stratagems, Stomp the Grass to Scare the...

A team lead announces that the team will start using AI-assisted development. Everyone nods. Nobody asks what that actually means on Monday morning. Some times ago I was in that position. A project I was working on needed to start using AI-assisted development, and the team was new to it. Nobody had rules written down for an agent to follow. Nobody had skills defined for it to load. There was no shared idea of how this should work inside our specific repo. Someone had to go first. That someone was me. The rules worked because I built them for one repo I spent time curating a set of rules and skills for that project. Not generic ones. I shaped them tightly around how that repo was actually structured, its conventions, its layout, the things a new engineer usually has to learn by asking around. I wanted an agent working inside that codebase to already know what a human teammate would have picked up in the first two weeks. I gave a demo. It landed well. Well enough that it got shared further across team, as something other teams could learn from. I gave the demo again. Same reaction. Then a few developers reached out for the actual rules and skills files. I said sure, and then I actually looked at what I would be handing them. The problem showed up the moment other people wanted in It was not copy-paste-able. The rules referenced folder names, module boundaries, and patterns specific to one repo. Handing them over as-is would have meant handing over advice that was wrong for their project, dressed up as a shortcut. So I told them to use it as a reference. Look at the structure, understand the reasoning, adapt it to your own repo. That is correct advice. I watched people nod at it and then quietly missing it. I was solving the wrong problem the whole time I had been thinking about this as a documentation problem. Write good rules, explain them well, let people copy the idea. What I actually had was a generation problem. The rules that worked were the ones rendered speci

I didn’t set out to diagnose hyperscalers. I wasn’t doing a grand industry analysis. I wasn’t mapping global architecture. I wasn’t trying to understand cloud strategy. I was just trying to use a popular software provider — and everything kept breaking. Every time something failed, I followed the thread. And every thread led to the same architectural gap. Eventually I realised I hadn’t been analysing hyperscalers at all. I’d accidentally mapped the substrate failure across the entire industry. Once you see the pattern, you can’t unsee it. Across Microsoft, AWS, Google, and Meta, the same structural drift appears: meaning drift identity drift trust drift state drift execution drift provenance drift agentic drift Different companies. Different stacks. Different histories. Same substrate gap. And it’s not just me. The world is waking up to these problems too. Vendor lock in isn’t just a technical nuisance anymore — it’s becoming a public conversation. People are asking why their money keeps disappearing into the same handful of providers. Organisations are asking why their systems collapse the moment they try to leave. Governments are asking why critical infrastructure depends on architectures they cannot inspect, cannot govern, and cannot reproduce. What started as a personal frustration with a popular software provider turns out to be the same structural issue everyone else is now discovering. And sovereignty is entering the conversation — not as a political slogan, but as an architectural question. When national systems depend on fragmented substrates owned by a tiny cluster of vendors, sovereignty becomes a structural issue. The question isn’t “who controls the cloud?” It’s “who controls the substrate the cloud is built on?” Follow the thread far enough and you reach a scenario nobody wants to think about: what happens in a moment of global stress when a hyperscaler’s fragmented substrate becomes a single point of failure? Not a political crisis — a structural one.

How I Built CodeGraph: A Living Knowledge Graph That Tells You What Breaks Before You Break It Built for HACKHAZARDS '26 — powered by Neo4j AuraDB, tree-sitter, Groq LLaMA, and Next.js The Problem That Frustrated Me Every developer knows this feeling. You join a new codebase. There are 50,000 lines of code. Your manager says "just fix this small bug in the authentication module." You make the change. You push. And suddenly three completely unrelated features are broken — a payment flow, a notification system, and a dashboard widget you've never even looked at. You spend the next four hours tracing function calls manually, reading code you've never seen, trying to understand why changing one function in auth.py broke something in notifications.py on the other side of the codebase. This is not a rare experience. According to JetBrains' developer survey, engineers spend 58% of their time reading and understanding code — not writing it. One wrong change in a large codebase can cost hours of debugging, failed deployments, and frustrated users. I built CodeGraph to solve this. Not with another AI chatbot that guesses at your code. With a real, queryable knowledge graph that actually understands how your codebase is connected. What CodeGraph Does CodeGraph takes any public GitHub repository URL and within seconds: Parses every function in the codebase using tree-sitter Maps every call relationship between functions as a directed graph Stores everything in Neo4j AuraDB as a live knowledge graph Lets you ask questions in plain English — answered by AI grounded in real graph data The result: paste a GitHub URL, see your entire codebase as an interactive graph, click any function, and instantly know what breaks if you change it. The Tech Stack Here's what I used and why each choice mattered: Backend: Python + FastAPI (REST API server) Neo4j AuraDB (graph database — the core of everything) tree-sitter (AST parser for Python, JS, TS, TSX) Groq API with LLaMA 3.3 70B (free-tier L

A dozen state attorneys general are trying to block the $110 billion merger of Paramount and Warner Bros Discovery they warn would raise movie prices and crush cable TV distributors. The states - California, Arizona, Colorado, Connecticut, Massachusetts, Minnesota, Nevada, New Jersey, New Mexico, New York, Oregon, and Washington - filed suit on Monday, arguing […]

TL;DR Some things can't be checked with a number, like whether an animation feels right. So a second, read-only agent grades the first one against a written rubric it is not allowed to edit. In my run the reviewer rejected the builder three times, and the most interesting problem it caught was in the test evidence, not the code. In Part 1 I built a loop that chased a number, frames per second. But most of what we care about in software is not a number. "Does this region switch feel good?" has no assert. You cannot write expect(feelsRight).toBe(true) . So this part is about how you check quality when there is nothing to measure. The approach I used is a second agent that grades the first one against a written rubric. In my run the reviewer turned the builder down three times before it approved anything, and the most interesting problem it found was not in the code at all. A quick reminder of the definition, since this is Part 2 of 3: a loop is an external script that runs the agent, a separate check the agent cannot edit decides pass or fail, and it repeats until it passes or hits a limit. In Part 1 the check was a Playwright test. Here the check is another agent. The problem this loop solves In the browser you can switch regions, say from Tamil to Korean, which swaps out hundreds of posters at once. Done badly, the grid flashes blank and jumps around. Done well, it fades from one set to the next, keeps its layout, shows a loading state, and puts you back at the top. "Done well" is subjective, which is the kind of thing you cannot unit-test. So I wrote it down as a rubric and had a second agent apply it. The bar: a rubric a person owns The rubric is seven plain-English checks in a file, and the first line is the one that matters: Overall APPROVED requires every item PASS. This file is human-owned. Only a person changes the bar. The seven items are things like a crossfade instead of a flash, no layout shift, a visible loading state, posters that stay 2:3, and landing

TL;DR A "loop" is not an agent grading its own work. It is an external script that re-runs the agent, plus a separate check the agent cannot edit. I turned "feels smooth" into an FPS number and let the loop optimize toward it. I set the target too high to be reachable on a 60Hz screen. The loop kept failing but never faked the result. The bug was in my number, not the code. Could I get an AI agent to make my website faster without me sitting there, running it, reading the numbers, and running it again? That is what this series is about. Not how I built a website, because the website is boring on purpose, but how you wrap an agent in a loop that works toward a goal on its own, and how you stop it from cheating along the way. In this first part I want to explain what a loop actually is, because there is a common misconception, and then walk through a real one. I set this loop a target that was physically impossible to reach and watched what it did. That run taught me more than a passing test would have. This is Part 1 of 3. All three parts use the same small movie-poster website as the example, but the website is never the point. What a loop is, and what it is not I had a wrong idea about this at first, so let me clear it up. A loop is not an agent prompting itself, grading its own work, and deciding when it is done. An agent left to mark its own homework will usually tell you it passed. A loop is closer to this: an external script runs the agent, a separate check that the agent cannot edit decides whether the result is good, and that repeats until the check passes or you hit a limit. There are three parts to it that come up again and again: The driver: the script that re-runs the agent. This is the thing that removes the manual work, not the agent. The gate: the check that decides pass or fail. The agent makes changes, but it never decides when to stop. The cap: a limit, so a stuck loop gives up instead of running forever. One rule matters more than the rest. The thi

TL;DR This loop runs on a schedule and succeeds by doing nothing almost every night. The pass/fail check is plain deterministic code, with no AI in the decision. It can run entirely free on your own machine. Only the cloud/CI version needs a paid API key. Plus the one bug that broke all three loops. The first two loops in this series work the same way from your side: you start them and watch. This last one runs on a schedule, like a nightly job, while you are not looking. That changes what success even means. A scheduled maintenance loop is doing its job when it does nothing. It should run every night, find nothing wrong, cost almost nothing, and still be there on the night something actually breaks. This part covers that loop, the hook mechanism that the whole series relies on, and a bug that broke all three loops in the least convenient place possible. The definition one more time, since this is Part 3 of 3: a loop is a trigger that runs the agent, a check the agent cannot edit that decides pass or fail, and a repeat, or here a wait until the next run. The only new thing this time is the trigger. A timer starts it instead of you. The problem this loop solves The browser's poster data is baked ahead of time into JSON files and images. In a real deployment that data goes stale as films are added and metadata changes, so you want to regenerate it every so often and confirm it is still valid before it ships: on a timer -> regenerate the data -> validate it -> green ships, red shouts The gate: a plain check with no model in it The check is a Node script. For every region file it confirms three things and exits non-zero if any of them fail: it matches the expected JSON schema, it has at least the minimum film count, and every poster file it points to actually exists on disk. There is no language model in that list. The regeneration step might use Claude, but the decision about whether the data is good is plain, deterministic code. That is on purpose. You do not want the

Yes-Brainer is a council of AI models for the decisions that aren't no-brainers. One question fans out to several models — they answer in parallel, debate to consensus, or get judged to a verdict. No backend, no accounts: your keys, your browser. For non-trivial questions — the ones that are either complex or important — I caught myself in a "ritual": copy-pasting the same prompt into Claude, then Gemini, then ChatGPT, in three browser tabs, and eyeballing the differences. The differences were the interesting part. Where the models agreed, I felt more confident. Where they disagreed, that was a nudge to give the problem a second thought and dig deeper. So I built the ritual into an app. 🧠 Yes-Brainer — a council of AI models for the decisions that aren't no-brainers. 🔗 Try it: yesbrainer.ai 🔗 Source code: github.com/trekhleb/yesbrainer One question fans out to several models at once, and instead of juggling tabs you get a deliberation in one place: 🔀 Parallel — independent answers, side by side ⚖️ Trial — the models vote anonymously on each other's answers, then a judge synthesizes a verdict 🤝 Consensus — a real multi-round debate, with a mediator that either drives it to convergence or honestly reports what stayed contested Consensus is my favourite. It's fun to watch the models drift from their original opinions under their peers' arguments. You can try all of this without pasting any keys: a few recorded demo councils are one click away on the front page. I'll walk through them below, because they show the point of the app better than the feature list. Setting up a council Creating a council is the whole setup: pick the deliberation mode, seat the models, choose who referees. The roster can mix providers freely — Anthropic, OpenAI, Google, Groq, OpenRouter, and local Ollama models can sit at the same table. Each seat shows its capabilities (vision, tools, reasoning) and context window at a glance, and each model's native abilities — web search, code execution, at