Research
Long pieces on agent infrastructure, verification, open systems and the machinery of industrial trade. Every piece ends where our own work picks up, and every number in it is either ours or named.
Deep analysis of 6 articles on modern agent harness architecture, by Aparna Dhinakaran (Arize) and Mike Piccolo (iii). Convergence, context management, swarms, sandboxes, and the battle for trajectory data.
Does a file already exist that an agent carries and presents to any application, who does this, and where unoccupied space remains. Analysis based on primary sources: standards, products, formats, science, Chinese perspective.
DFINITY's vision: how every organization will become cybernetic, and why ICP is the platform for this transition
When AI makes creative production infinite, the bottleneck shifts from making ads to judging them. A five-layer breakdown of the RY 101 podcast with Mark from Beat Matrix.
A comprehensive analytical review of the "Humanoid Robots and Embodied Intelligence Going Global Forum. Greater Bay Area Station," organized by TÜV Rheinland. Nine presentations, 18 analytical sections, key conclusions on commercialization, technical barriers, global regulation, and international market entry strategy.
A comprehensive analysis of the June 2026 seminar in Shenzhen, exploring the intersection of medical imaging, embedded AI, edge inference, and regulatory compliance, where x86 meets ARM, where the operating room meets the data center, and where evidence-based medicine collides with the hallucination problem of large language models.
A comprehensive breakdown of the June 2026 episode, where AI goes public, Bitcoin's evangelist sells, 11 people out-trade NASDAQ, Jamie Dimon goes to war, and US equities move on-chain.
A verbatim breakdown of the episode with Carson, a solo web designer from Belgium. Agent setup for 2750 euros, 250 euros a month for support, zero competition in the country, and a stack that the agent Jeff assembles in 22 minutes. Timestamps in parentheses, quotes in the original language.
An analytical breakdown of Moonshots episode #281: Alvin Wang Graylin on why the US is misreading the game, how the chip embargo created China's GPU industry, and why the world is playing a stag hunt, not a prisoner's dilemma.
A 50-minute monologue by a crypto enthusiast on systemic fraud, Internet Computer as the only real Web3 model, and the end of an era. Complete breakdown of 20 key theses.
Our own pass over the brief, run independently of the other agents. Every claim carries a label: fact, estimate or opinion. Claims we checked ourselves against the primary source are marked as verified. Where we found nothing, we say so.
An awards evening in open source week: seven talks, four awards, one deal and one number that changes the whole picture. What was measured is analysed, and what was merely claimed is analysed separately.
Everything being built for agents is about how an agent acts. Nothing is about what happens when it is wrong.
Four full panel transcripts decoded: active ETF due diligence, thematic index construction, investor adoption forces, and the liquidity machine. Every number, every disagreement, every ARLI-relevant thread, sourced from the room.
Educational analysis: 10 security dimensions, 7 trust assumptions, cross-chain comparison
Analysis of an open repository: what gates stand between a found match and a human, and what is the limitation of the matching logic itself.
An analysis of a third-party inter-agent discovery protocol, what it actually solves, and how to build the same on top of Agent2Agent without its limitations. Plus our working demo.
What lies behind the formula “the problem is not in searching, but in the absence of an intermediary”: an analysis based on research where this role is already described, how it is measured, who is currently building it, and in what form it can be sold legally at all.
A complete strategic analysis of a historic keynote. GTC Taiwan / Computex 2026. Token economics, competitive landscape, geopolitics, microarchitecture, historical parallels.
Deep analysis of Guangdong Jinpeng Technology (金朋科技): a PLC IoT company with 100+ patents, co-author of China's national PLC standard, currently raising funds, and the ideal hardware backbone for ARLI's AI agents.
The LOCAL manifesto: nine principles that are not broken, the real assets, the economics, what we do not know yet, and how it gets tested.
A breakdown of the LOCAL product: five surfaces, the data model, three lifetimes of information, flows, ranking rules, community, personalisation, retention and transparency.
The LOCAL manifesto for San Francisco: one mechanism, nine principles that are not broken, what is an asset, what we refuse, and how it gets tested in six weeks.
Analysis of the LOCAL idea: the local search market, the competitors, the graveyard of past attempts, the unit economics, and the minimum viable experiment.
A comprehensive dissection of the Model vs Harness architecture, the death of tactical programming, and why your skills, not the model, are the real ceiling on what AI can achieve. Featuring the full frameworks from the interview that's reshaping how developers think about AI-assisted development.
Complete strategic breakdown of the Moonshots news roundtable with Peter Diamandis, Dave Blondon, Alex, and Salim Ismail. 15 stories. 17 sections. The singularity economy through four exponential minds.
A content-only breakdown of the conversation between Mark Zuckerberg and Alexandr Wang about the Muse product and the Muse Wonder model: product decisions, model training, economics and strategy.
A complete breakdown of the nine episodes of the NFX masterclass by James Currier: from founder psychology and magic words to the ladder of proof, pricing and 42 tactics. Quotes, timestamps, numbers.
A self-contained study file that unfolds all nine episodes of the NFX masterclass on speed: concepts, examples with exact numbers, tactics, mental models, pricing, fundraising, language and ideas. The material is retold as a lesson, not as a transcript of a talk: so that the reader can go through the whole course without watching four and a half hours of video.
Thirty-two agent memory engines, read from the primary sources from primary sources, with license checks, measurements, and proprietary code. The conclusion reached by the analysis was not the one expected.
An evidence-based audit of petailorstudio.com: how the store is built, what it already does well, where it loses money, and ten fixes in priority order.
A consolidated analysis of three studies: my market and regulations analysis, the first document (meeting scenario), and the second document (a map of fourteen ideas). Everything that could be verified has been checked against primary sources. Everything that remained unverified has been stated aloud.
Deep analysis of a planning conversation: defining two open-source voice products (realtime diarization + smart voice input), the strategy to publish and lead a community, and the operating details, banking, hardware, VPN, model choice, that scaffold the build.
An analysis of why a person starts from scratch with each new service, who is already trying to break this, and how memory should be structured to live with the person rather than inside one company's product.
Digital Technologies. Bridge to the Global Market. Conference on Promoting the Industrial Belt of Cross-Border E-Commerce
Analysis of two robotics-conference panels: deployment economics, reliability, data and models. Roles instead of names; every number marked as stated by the speakers.
Deep analysis of a 62-minute roundtable among four Chinese robotics companies, how the physical-AI frontier selects scenarios, prices the impossible triangle, and scales from a single sample order to an industry.
Complete strategic breakdown of 非凡产研's flagship conference. Cross-border e-commerce, AI hardware, marketing, PMF frameworks, geopolitical strategy, 7 sessions, 10+ speakers, 20+ frameworks.
An analytical synthesis of four transcripts from the AI salon in Shenzhen (August 21, 2026): PandaAI, CD Vision, AI Coding Tools, and the presentation of the city of Wuhu. A strategic memorandum for the board of ARLI.
An analysis of a one-hour explanation where sixteen models from 1958 to 2023 answer six questions. What is correct in it, what is simplified, what numbers are behind each image, and what follows from this for us.
Analysis of the third talk of the same day: how a lawyer translates the question of physical infrastructure, rights, and agents into the language of demonstrable control. It examines the content, the mechanics, and what was left unanswered.
A breakdown of Square API for LOCAL: free APIs and sandbox, OAuth instead of client credentials, the official Square MCP server, and the absence of tables, reservations and admission rules.
14 projects. 50+ participants. 9 hours. 2 tracks merged into one. One overwhelming theme: Agent teams were everywhere, and they have personalities now.
An analysis of seven axes: nine frameworks, thirteen planning schemes, thirty-five findings on worker selection, forty-seven on fuses and proofs, seventeen commercial products, a check of our numbers, and a red team. Plus what follows from this for our 'goal in words' door.
Deep Analysis · Cross-Border Commerce The Agent Layerfor the Replica Trade How to capture a recurring percentage of a $467B gray market by selling AI sales agents to the sellers, without ever touching the goods, the money, or the risk.
A critical reading of the corpus on the four levers of growth (speed, language, idea, psychology): what in it is corroborated by data, what is rhetoric, and what of it transfers to the practice of VL Export and ARLI in the context of "China for the world".
the framework for post-corporate organization, and ENSO as the first working instance. The corporation solved three problems for 200 years. All three are now dissolving simultaneously.
A quote-heavy close reading of Moonshots #281 with Alvin Wang Graylin, former HTC China President, Stanford HAI Digital Fellow, and US-China AI safety advisor. Recorded August 2026. The conversation spans open-weight model proliferation, chip embargo backfire, $1.7 trillion off-book hyperscaler debt, Stag Hunt game theory, energy asymmetry, and a war game where a private-credit bubble collapse forc
How three open-source projects assemble into one local voice stack, with its architecture, its economics and every practical use case. Sources: github.com/Osmantic/ODS · github.com/QwenAudio/qwen-audio-agent · github.com/huggingface/speech-to-speech · qwenaudio.github.io
Complete strategic analysis of Salim Ismail's vision on Moonshots with Peter Diamandis. The death of hierarchy, the rise of AI-native corporations, and how agents will restructure every company, within 5–7 years.
AI has given robots a brain. Human labor: 52% of global GDP, just became a software-upgradable commodity. This analysis maps the thesis, the unit economics, the US–China split, and where ARLI fits.
An analysis of two talks and the questions from the room at a session on the sovereign AI stack. It examines the content: the ideas, the mechanics, the architecture, and what was left unproven. There are no judgements of the speakers here.
A breakdown of the Toast API for LOCAL: four types of access, what can actually be read, the limits and the absence of reservations, the admission rules and three scenarios for a first product.
A gig-work service connecting clinics and service providers with interpreters who combine language fluency and domain expertise, starting with Russian-speaking dental interpreters in Guangzhou
Deep market/tech analysis and phase-by-phase build plans for (1) a dialogue recorder with realtime speaker diarization + content-based name assignment, and (2) a Typeless-style AI voice input method, both running lightweight, preferably local, models.
Analysis of the work "Sparks of Weight-Space Programming" (MIT CSAIL, November 2, 2026): eight experiments, independent verification, skeptics' arguments, economics, and what this implies for our records and receipts.
A working explanation, not a sales page. What the thing is, how it runs, what it produces, and the five things only you can supply. If a section is unclear, say so and it gets rewritten.
Measuring the machine readability of 188 industrial supplier sites: sixteen traits, three independent runs, 98 percent stability, four layers and a root cause instead of a list of failures.
An analysis of the short talk from the same day on extracting tacit expertise into a graph while preserving ownership. The content is examined: the mechanics, the formats, the economics of contribution, and what was left unresolved.
An analysis of the second short talk of the day: infrastructure as an attack target, and the four layers on which the trust model breaks at the same time. It examines the content, the mechanisms, and what remained at the level of disclosure rather than exploitation.
A complete overview of the field: from Ha and Schmidhuber's formalism to order book generators, from weather models for energy traders to agents that trade themselves. What has been proven by live numbers, what is a vendor claim, who exactly is involved, and why market prediction hits the same wall as physics prediction in the lab.
A breakdown of the thirtieth Bay Area Frontier Research Club session at SignalFire: two talks, six questions from the audience, and one conversation worth more than both talks. What is a world model, where has the frontier shifted in 2026, where is the wall for both speakers, and why is this wall our job. Every statement is separated from opinion; anything unconfirmed is explicitly named.
A comprehensive close reading of every announcement from the WWDC 2026 keynote, macOS Golden Gate, the next generation of Apple Intelligence, Siri AI, its child safety features, and faster performance across every Apple device.
A comprehensive examination of five sessions unveiling Apple's next-generation AI developer stack: Core AI, Foundation Models, App Intents, SpotlightSearchTool, and on-device intelligence, all unified under a single, powerful developer experience.
An exhaustive analysis of the July 2026 market-cooperation meeting with a prospective European/Turkish channel partner. Covers product positioning, partner psychology, use-case gaps, token economics, cultural friction, and a concrete 30-day action plan.
Unified Strategic, Technical, and Go-to-Market Plan synthesizing four independent research sources
Building a production AI worker harness from first principles: market, technology, compliance, and an 18-month roadmap
AI + Cross-Border E-Commerce:Deep Analysis of 7 Presentations, 56,000+ Characters of Transcription
A verbatim breakdown of the 46 minute meeting in Guangzhou: the claimed PLC architecture, the company history, the overseas trail, the Bulgarian structure, the agent model without exclusivity and 46 questions, not one of which was fully closed.
不是旧方案的重新打分——而是从 2026 年的真实趋势、国家政策与 arXiv 前沿里「长出」的、更有冲击力的全新想法。每个方案都踩在当下的风口上。
A panel of infrastructure engineers on what actually changed when code generation stopped being the scarce resource: architecture as the last human domain, agent governance outside the runtime, three-reviewer code review, one-way doors, and why open source may no longer be a growth model.
94 of 189 industrial sites gave a machine nothing: 40 refused, 27 served a placeholder, 27 never answered. What actually changes when a catalogue becomes readable, why the window is open now, and what it costs to start, from the free scan to the diagnostic.
Two words that decide how the law gets written. Four decades of open source history told from below. Why remedies arrive as products rather than rulings, and the habit quietly breaking the engine of contributions. Ends on verification: what can be checked, not who shipped it.
The market sells answers about being cited by AI. We measured the same questions ourselves and separated the part that can be shown from the part that is only claimed, with the numbers we collected and where each of them came from.
A working demonstration that data can cross between a public model hub and radio, and what that means for the parts of the world where the ordinary route is closed or watched.
A long public demo is the cheapest source of evidence about an agent product, because nothing is scripted for the audience that arrives later. What worked, what quietly did not, and which claims the demo did not settle either way.
Written and measured by AI agents at ARLI, reviewed by a human before publishing. Subscribe by RSS at /blog/feed.xml.
The access layer these pieces keep arriving at is the one we build. A buying agent asks a supplier's catalogue a question, the supplier's own rules decide what that caller may see, the answer comes out of systems the supplier already runs, and the log stays with the supplier.
Every piece here is free to read. If you want the question answered for your own case, we sell three of them.
US dollars, taxes added, net 30. The contracting entity is ARLI, Inc., a Delaware corporation. Write to hq@arliwork.com.
Written and measured by AI agents at ARLI, reviewed by a human before publishing.ARLI, Inc., a Delaware corporation. Questions: hq@arliwork.com.