Cycle on Jun 14, 2026

I will start by listing the active threads and reading the existing notes for this topic. This will help me understand what research has already been conducted, identify any open questions or watches, and see which finding slugs I should reuse or update.

I will read the existing notes for the topic to see what has already been documented regarding Sacra, Google Scholar, Perplexity, Gemini Deep Research, and the market map.

I see several prior findings. I also have some open threads/watches. Looking at the current Hacker News stories and themes: Theme 6 has a story: - `[592 pts] Amazon CEO's talks with U.S. officials triggered crackdown on Anthropic models — https://news.ycombinator.com/item?id=48519092` This is highly relevant to our watch: - `Anthropic Restores Access to Fable 5 and Mythos 5` (id: `019ebf48-b349-73e7-984e-261a5c8bc427`) And our note: - `claude-fable-5-silent-safeguards-and-pricing-shift` Let's inspect the watch: "Monitor when the US government lifts the export control directive on Fable 5 and Mythos 5, or Anthropic officially restores access to these models for users." Let's search Hacker News or look at this story in detail. Let's run `hn_story` on `48519092` to see what is going on with the Amazon CEO talks and the crackdown on Anthropic models. This looks extremely relevant to the watch and the note `claude-fable-5-silent-safeguards-and-pricing-shift`. Let's do that.

This is a major piece of news that directly impacts our watch `Anthropic Restores Access to Fable 5 and Mythos 5` (id: `019ebf48-b349-73e7-984e-261a5c8bc427`) and our existing note `claude-fable-5-silent-safeguards-and-pricing-shift`. Let's look at the details of the news: - Amazon CEO Andy Jassy had talks with Trump administration officials (including Treasury Secretary Scott Bessent). - Amazon researchers used prompts to get Anthropic's Fable 5 model to provide information that could aid cyberattacks. - White House officials held a meeting and security researchers tested Amazon's claims, leading to a sweeping move preventing foreign governments, companies, and individuals from accessing the tool. - Anthropic responded by shutting off access to Mythos and Fable to *all* users (effectively, a complete shutdown of foreign/domestic access to prevent violating the ban or due to foreign-born employees being unable to work on it). - This is part of a broader conflict between Anthropic and the Trump administration, which designated Anthropic a security risk (Pentagon) and has been fighting in lawsuits. - Jassy's calls escalated into a wide Commerce Department ban on foreign users. This is extremely relevant to our notes on AI capability, safety, and supply chains. Wait, let's look at the other stories/themes. Let's do a search on Hacker News for Sacra, Perplexity, Gemini Deep Research, and others. Let's see if there have been any new developments on Sacra, Perplexity, Google Scholar, Google Deep Research, or OpenAI Deep Research. Let's search for "Sacra" first using `hn_search`.

The keyword search "Sacra" returned mostly Sacramento. Let's search for "Sacra research" or "sacra.com" instead. Let's run a web search for Sacra research to see how they are doing and what they do. Let's search: `Sacra research platform AI` or `Sacra research API` or `sacra.com` via `web_search`.

Okay, let's search specifically for what Sacra does. Let's do `web_search` with `site:sacra.com` or `"Sacra" "private market research"` to understand its model. Let's search `sacra.com research model pricing` or `site:sacra.com "about"`.

Let's check `https://sacra.com/` and other pages to understand Sacra's features, value proposition, technology, data source, and pricing model. Let's do `web_search` with `"Sacra" pricing model` or `"Sacra" business model` or `"Sacra" "pricing"`.

Let's fetch `https://sacra.com/pricing/` to see their subscription tiers and details. Let's do that!

This is extremely clear and detailed. Let's analyze Sacra's pricing model: - Monthly Standard: $50/mo (1 seat, 500 tasks included, company & market reports, charts & datasets, expert interviews, Model Context Protocol (MCP) server access, API, community, revenue & valuation models, Q&A, filings, XLS & PDF export, Bespoke research). - Pro: $350/mo (5 seats, 3500 tasks included). - Platform: $1500/mo (20 seats, 25000 tasks included). - Enterprise: Custom pricing. Wait, what is a "task" in Sacra? Let's check: "tasks included". This represents programmatic agentic tasks, maybe API calls or MCP server queries? Let's see if we can find more about Sacra's core technology, "Sacra Agent", or "Sacra MCP". Let's do a search on LinkedIn or Twitter or general web for: `"Sacra" "MCP" OR "Model Context Protocol"` or `"Sacra Agent"`. Let's search: `Sacra "MCP" OR "Model Context Protocol"` via `web_search`.

Let's fetch `https://sacra.com/features/mcp/` and `https://docs.sacra.com/` to understand how Sacra integrates private market research as agent-ready infrastructure. Let's fetch `https://sacra.com/features/mcp/` first.

Let's fetch `https://docs.sacra.com/` to understand what they do. Wait, does `https://docs.sacra.com/` load? Let's check. Let's do `web_fetch` for `https://docs.sacra.com/`.

This is excellent. Let's see what else we can find about Sacra, Perplexity, Gemini Deep Research, Google Scholar, and OpenAI Deep Research. Let's look at the prior findings to see what we've already written. Our prior findings: - `market-map-positioning-hey-lefty`: "The landscape of autonomous and AI-powered research tools has matured into a highly stratified market in 2026. Rather than a single monolithic category, the market is divided into distinct tiers…" - `perplexity-deep-research-consumer-agent`: "Perplexity has solidified its position in the autonomous research market by offering a highly optimized, high-speed consumer search and synthesis engine. Rather than focusing on long-running…" - `academic-ai-research-scholar-vs-synthesis`: "The academic and scientific research ecosystem has evolved beyond simple keyword searches (like Google Scholar) and generic chat models. In 2026, the market for specialized academic AI research is…" - `integrating-custom-mcp-servers`: "To realize its positioning as a model-agnostic, multi-source research orchestrator, Hey, Lefty can natively implement support for the Model Context…" - `gemini-deep-research-agent`: "In April 2026, Google DeepMind announced a major evolution of its autonomous research agent ecosystem with the release of Deep Research and Deep Research Max in public preview on the Gemini…" - `sacra-premium-private-market-mcp-data`: "Sacra has aggressively pivoted to position its private market intelligence as "agent-ready infrastructure" by unbundling its research reports, financials, and company profiles into programmatic…" - `institutional-mcp-financial-data-servers`: "A major structural shift in the financial intelligence market has solidified: institutional data giants have officially unbundled their massive, proprietary data networks into agent-ready,…" - `openai-deep-research-api-and-pro-tier`: "OpenAI's programmatic deep research capabilities have undergone a rapid structural transition. After releasing dedicated deep research models in the API—namely o3-deep-research and…" Wait! The user wants us to help build a market map to see where "Hey, Lefty" fits and how to position it. Let's review the details of the competitors: 1. **Gemini Deep Research (Google DeepMind)** - What it does: Multi-step, long-running research agent that can execute dozens of steps, formulate search queries, read web pages, PDFs, and write comprehensive, structured reports. - Core technology: Built on Gemini 1.5 Pro / Ultra with the Interactions API (multi-step reasoning, tool-use loops, native multimodal processing). - Data & Sources: Google Search index, Google Scholar, Web index, academic papers, and uploaded files. - Target Customer: Knowledge workers, researchers, analysts, enterprise teams. - Pricing/Business Model: Part of Gemini Advanced ($20/mo) or via the Gemini API (Interactions API billing / pay-per-token or specialized pay-per-run). 2. **Perplexity Deep Research** - What it does: Consumer-centric, high-speed autonomous synthesis. Generates follow-up questions, plans a multi-step search, reads multiple web pages simultaneously, and synthesizes a structured summary. - Core technology: Built on top of their custom search infrastructure, integrated with models like Claude, GPT, or Sonar. Uses parallel web searching and fast RAG. - Data & Sources: Live web search, news APIs, and integration with academic databases. - Target Customer: Everyday consumers, professionals needing quick overviews, students, general knowledge seekers. - Pricing/Business Model: Perplexity Pro ($20/mo), with API access charged per token/query. 3. **Google Scholar / Academic AI Research (Consensus, Elicit, Scite)** - What it does: Google Scholar is a traditional search engine for academic literature (keyword-based, citation indexing). Consensus, Elicit, and Scite are specialized AI academic research tools that extract findings, synthesize literature reviews, evaluate study designs, and map citation networks. - Core technology: Specialized semantic search, vector embeddings of academic databases (Semantic Scholar, PubMed), LLM-based metadata extraction and synthesis. - Data & Sources: Academic publishers, Open Access repositories, Semantic Scholar API, PubMed, Crossref. - Target Customer: Academics, researchers, scientists, clinicians, students. - Pricing/Business Model: Elicit / Consensus use freemium models with subscription tiers ($12-$50/month) based on credits/runs. Google Scholar remains free (ad/search-supported). 4. **Sacra** - What it does: Private market research platform. Provides company reports, analyst interviews, revenue models, and curated news on growth-stage and pre-IPO companies. - Core technology: Unbundled its data into programmatic endpoints and MCP (Model Context Protocol) servers. Allows AI agents (ChatGPT, Claude, custom agents) to query financial models, valuations, and expert transcripts directly. - Data & Sources: Proprietary analyst research, expert interviews, private company filings, and cap table models. - Target Customer: Institutional investors (VCS, PE, hedge funds), secondary market buyers, startup founders, and platform developers building AI investment tools. - Pricing/Business Model: Subscription model: Standard ($50/mo), Pro ($350/mo), Platform ($1500/mo), and Enterprise (Custom). Usage is managed through "tasks" (e.g., 500 tasks, 3500 tasks, 25000 tasks). 5. **OpenAI Deep Research (o3-deep-research, etc.)** - What it does: Programmatic deep research models integrated into ChatGPT Pro and available via API. Focuses on complex reasoning, coding, and multi-step research tasks. - Core technology: o3 and o3-mini reasoning models with deep search capabilities. - Data & Sources: Web search index, partner data, system integration. - Target Customer: Developers, power users, enterprise researchers. - Pricing/Business Model: ChatGPT Pro ($200/mo) or API pay-per-token/run. Let's look at **Hey, Lefty**'s positioning: Where does **Hey, Lefty** fit? Hey, Lefty is a model-agnostic, multi-source research orchestrator. Instead of being locked into a single model provider (like Gemini Deep Research or OpenAI Deep Research) or restricted to a single closed data index (like Perplexity or Sacra), Hey, Lefty orchestrates multiple search engines, local and cloud LLMs, and custom programmatic data sources (such as custom MCP servers from Sacra, FactSet, PitchBook, or internal company databases). This allows Hey, Lefty to act as the "control plane" or "orchestration layer" for professional research. It coordinates the research cycle, writes findings, monitors dated catalysts/watches, and integrates custom MCP servers. Let's do a search on other emerging research tools and platforms in 2026 to see if there are other players we should map. Let's search: `autonomous research agent AI market map 2026` or `AI research tools market map` or `financial research AI agents` via `web_search`.

Let's fetch `https://mightybot.ai/blog/ai-automation-agents-market-maps-gone-wild/` to see what they say about AI research agents and how the market is categorized in 2026. Let's do `web_fetch`.

The fetch failed due to a timeout. No problem, let's try a different search or fetch. Let's do a search on: `"MightyBot" "AI Agents Market Map 2026"` or let's search for `"research agent" market map` or `"autonomous research" market map`. Let's search: `"autonomous research" agent market map OR categories OR segments` using `web_search`.

Let's check the paper `Deep Research Agents: A Systematic Examination And Roadmap` (published around 2025/2026). Let's do a `web_search` for `"Deep Research Agents: A Systematic Examination And Roadmap"` or search for its abstract or markdown. Let's search: `"Deep Research Agents: A Systematic Examination And Roadmap"` via `web_search`.

Let's fetch `https://medium.com/@sulbhajain/deep-research-agents-paper-review-f2904292c43e` or `https://www.linkedin.com/pulse/deep-research-agents-systematic-examination-roadmap-rafael-pel%C3%A1ez-jf2qf` to understand how the paper categorizes Deep Research Agents. Let's do a `web_fetch` on `https://medium.com/@sulbhajain/deep-research-agents-paper-review-f2904292c43e`.

This is an extremely valuable academic review. Let's analyze the core concepts mentioned in the review of the "Deep Research Agents: A Systematic Examination and Roadmap" paper (arXiv:2506.18096, cs.AI): 1. **Definition of Deep Research (DR) Agents:** - Advanced AI systems built on LLMs that autonomously tackle complex, multi-step research tasks. - They dynamically reason and adapt their plans over long, multi-turn sessions. - They retrieve information from diverse sources in real time (APIs and web browsing). - They use external tools iteratively (code execution, multimodal data processing). - They produce structured analytical reports that synthesize findings and insights. - Unlike traditional RAG (which lacks deep reasoning) or conventional Tool Use (which follows rigid workflows), DR agents offer autonomy, continual reasoning, and adaptive planning. 2. **Key Tech Components:** - **Information Acquisition:** Contrasts API-based retrieval (high data quality/structure but restricted scope) with browser-based exploration (high coverage but unstructured) to maximize knowledge coverage. - **Modular Tool Use:** Integrates code execution, multimodal inputs, and standardized protocols like the Model Context Protocol (MCP) for extensibility. - **Workflow Design:** Static vs. dynamic, single-agent vs. multi-agent. - **Optimization:** Prompt engineering, fine-tuning, reinforcement learning, and non-parametric continual learning. 3. **Ecosystem & Industry Context:** - Major players: OpenAI, Google, Anthropic, Perplexity. - Collaboration protocols mentioned: - **Model Context Protocol (MCP)**: For tool and data integration. - **Google's A2A (Agent-to-Agent)**: For agent-to-agent communication. This is a fantastic framework to structure our Market Map and Positioning finding! Let's do a quick search regarding "Google's A2A" or "Agent-to-Agent" protocol to see what that is, as it's mentioned in the review. Let's search: `Google "A2A" OR "Agent-to-Agent" protocol AI` or `Google DeepMind "A2A"` via `web_search`.

This is a critical insight. We have two competing/complementary protocol standards in the AI agent space: - **Model Context Protocol (MCP)** (originally introduced by Anthropic) which focuses on connecting a model to local or remote data sources and tools (vertical: model to tools/data). - **Agent-to-Agent (A2A) Protocol** (introduced by Google DeepMind) which focuses on enabling interoperability and communication between different AI agents across frameworks, platforms, and clouds (horizontal: agent-to-agent). Let's fetch `https://www.koyeb.com/blog/a2a-and-mcp-start-of-the-ai-agent-protocol-wars` or search for it to get a clearer picture of A2A vs MCP. Let's do `web_search` for `"A2A and MCP: Start of the AI Agent Protocol Wars"` or fetch it. Let's do `web_search` first.

Let's fetch `https://www.merge.dev/blog/mcp-vs-a2a` or `https://heidloff.net/article/mcp-acp-a2a-agent-protocols/` to get the core details of these protocols. This will represent a massive addition to our mapping. Let's do `web_fetch` for `https://heidloff.net/article/mcp-acp-a2a-agent-protocols/`.

This is an incredibly rich set of insights. Let's summarize the protocols and how they map out: 1. **Model Context Protocol (MCP)**: - Creator/Sponsor: Anthropic. - Purpose: Standardizes how models connect to tools and data sources. It is highly popular due to its simplicity, extensive documentation, and integration with IDEs (like Cursor) and Claude. - Design: Originally intended for single-session environments (model-to-tool) rather than collaborative agent-to-agent architectures. - Limitations: Lacks built-in authorization (though being worked on), streaming across agents, persistent memory, and asynchronous long-running task execution. 2. **Agent Communication Protocol (ACP)**: - Creator/Sponsor: IBM (donated to the Linux Foundation). - Purpose: Standardizes agent-to-agent and agent-to-client invocation. It is designed to act as a "Swagger for agents." - Design: Addresses the gaps in MCP (streaming, memory, and asynchronous execution). It natively supports long-running tasks, persistent state/memory across multiple sessions, and "Human-in-the-Loop" confirmations where an agent can pause and await external responses. 3. **Agent-to-Agent (A2A) Protocol**: - Creator/Sponsor: Google DeepMind (recently donated to the Linux Foundation in late 2025). - Purpose: Focuses on horizontal agent-to-agent communication and interoperability across different platforms and cloud environments. - Design: Enables cross-framework, cross-organizational agent orchestration. Standardizes peer-to-peer collaboration, enabling a fleet of specialized agents from different vendors to coordinate. Let's do a quick search for Sandi Besen's article or other reviews to verify any additional technical details of A2A, ACP, and MCP. Let's do `web_search` for `Sandi Besen "Comparison of MCP, ACP, and A2A Protocols"` or `"Sandi Besen" "A2A"`.

Awesome. Let's fetch `https://research.ibm.com/blog/agent-communication-protocol-ai` to see their official description of ACP and how it contrasts with MCP and A2A. Let's do `web_fetch`.

This is an amazing explanation. It lays out how ACP and MCP are used together: - **MCP (Model Context Protocol)**: Connects a model/agent to its *tools* and *knowledge* (vertical integration). - **ACP (Agent Communication Protocol)**: Connects *agents to other agents* (horizontal integration, peer-to-peer delegation, memory transfer, streaming, async execution). - **A2A (Agent-to-Agent Protocol)**: Google DeepMind's open protocol for agent-to-agent interoperability across diverse frameworks and clouds (similar goal to ACP, but backed by Google and donated to the Linux Foundation in late 2025). Let's do a search on other elements of our brief: - What is the market for autonomous or AI research tools? - Features, value proposition, core tech, data source, pricing/business model of: - Gemini Deep Research (Google DeepMind) - Google Scholar vs specialized academic tools (Consensus, Elicit, Scite) - Perplexity Deep Research - Sacra (private market research) - OpenAI Deep Research (o3-deep-research) Wait, let's look at our previous notes on these. Let's see what we have on `market-map-positioning-hey-lefty` first. I will read it to see what was written and how we can refine it. Let's do `read_notes` and fetch the note `market-map-positioning-hey-lefty` if we can. Note bodies are retrieved by `read_notes` or we can just fetch it? Actually, `read_notes` returns a list of notes with their preview, but to read the full body, wait, does `read_notes` return the full body or is there another way? Wait, `read_notes` has a parameter `topic_id`. Let's check if we can read the note. Oh, wait, the `read_notes` description says: "List the current set of notes for a research topic. Returns slug, title, preview, updated_at, and source_count for each note so the cycle can reference prior findings without re-fetching them." Wait, does it return the body? If not, how do we read the full body? Let's check `read_notes` tool parameters. It has `limit` and `topic_id`. Let's call `read_notes` to see if it returns the full body, or if we can write to update it. Wait, the prompt says "Before anything else, read what's already in front of you: the 'Relevant prior findings' block...". In the kickoff message, we have: - `market-map-positioning-hey-lefty` - `perplexity-deep-research-consumer-agent` - `academic-ai-research-scholar-vs-synthesis` - `integrating-custom-mcp-servers` - `gemini-deep-research-agent` Let's review what was in the "Relevant prior findings" block of the user message: 1. `market-map-positioning-hey-lefty`: "The landscape of autonomous and AI-powered research tools has matured into a highly stratified market in 2026. Rather than a single monolithic category, the market is divided into distinct tiers…" 2. `perplexity-deep-research-consumer-agent`: "Perplexity has solidified its position in the autonomous research market by offering a highly optimized, high-speed consumer search and synthesis engine. Rather than focusing on long-running…" 3. `academic-ai-research-scholar-vs-synthesis`: "The academic and scientific research ecosystem has evolved beyond simple keyword searches (like Google Scholar) and generic chat models. In 2026, the market for specialized academic AI research is…" 4. `integrating-custom-mcp-servers`: "To realize its positioning as a model-agnostic, multi-source research orchestrator (see market-map-positioning-hey-lefty), Hey, Lefty can natively implement support for the Model Context…" 5. `gemini-deep-research-agent`: "In April 2026, Google DeepMind announced a major evolution of its autonomous research agent ecosystem with the release of Deep Research and Deep Research Max in public preview on the Gemini…" Wait, let's check what other notes we have: - `sacra-premium-private-market-mcp-data` - `institutional-mcp-financial-data-servers` - `openai-deep-research-api-and-pro-tier` - `claude-fable-5-silent-safeguards-and-pricing-shift` - `german-court-ruling-ai-overview-liability` - `agentic-search-grep-vs-vector-retrieval` Let's search for "Sacra" in our notes or retrieve its preview to see what is there. Wait, let's do a `web_search` for `Sacra MCP server` or `Sacra API` to make sure we have the latest. Let's see if we can read the exact notes using `read_notes` or if we should write updated notes. Let's call `read_notes` first to see if it returns the full body. Ah, `read_notes` returns slug, title, preview, updated_at, and source_count for each note. Let's check the output of the tool call `read_notes` we did earlier: It returned the list of notes with `preview`, `slug`, `source_count`, `title`, `updated_at`. Wait, can we read the full body? No, there is no separate `read_note` tool. But we can update notes using `write_note`. Let's assume we can write a comprehensive note on the market map and positioning of Hey, Lefty, and update the individual competitor profiles if we have new information. Let's look at the current events: 1. **The Claude Fable 5 Crackdown & Government Escalation**: - On June 12, 2026, the US government banned foreign users from accessing Claude Fable 5 and Mythos 5. - On June 13, 2026, news broke (via WSJ and HN) that this crackdown was triggered by Amazon CEO Andy Jassy's private talks with US officials, including Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick. - Amazon researchers used prompts to get Fable 5 to provide information that could aid cyberattacks. - In response to the foreign user ban, Anthropic shut down access to Mythos and Fable for *all* users (domestic and foreign) to ensure compliance and because their foreign-born researchers could not work on them. - This represents a massive political escalation and a huge "supply chain risk" for companies relying on a single centralized proprietary model (like Claude) for their autonomous research tools. - This provides a massive tailwind/justification for **model-agnostic orchestrators** like **Hey, Lefty**, which can dynamically route tasks to other models (like OpenAI's o3-deep-research, Gemini Deep Research, or open-weights models like GLM 5.2, MiniMax, or Z.ai) when one provider's models are restricted, banned, or silently degraded. Let's verify what GLM 5.2 is. Theme 1 mentions "GLM 5.2 Is Out". Let's run a `web_search` or `hn_story` search on GLM 5.2 to understand its capabilities and context. Let's do `hn_story` on `48518684`.

This is such a fascinating dynamic. Let's see: - GLM-5.2 is Zhipu's latest open-weight model with a 1M context window and leading performance on long-horizon tasks and coding. - It was released at 5:21 PM Chinese time, right around when Anthropic received the government letter banning Fable 5. Zhipu's founder (Jie Tang) explicitly framed the release as a geopolitical counter-move to Western model restrictions, saying "radical openness" is needed because "the path to AGI must never be enclosed by high walls" and "frontier intelligence must remain open-source, accessible, and buildable." - This highlighting of "radical openness" vs. "enclosed by high walls" is a direct reaction to the US government's export control directive on Anthropic's Fable 5 and Mythos 5. Wait, this is an incredible development for the geopolitical AI landscape, which directly intersects with our note `claude-fable-5-silent-safeguards-and-pricing-shift`. Let's review the threads/watches: - `Anthropic Restores Access to Fable 5 and Mythos 5` (id: `019ebf48-b349-73e7-984e-261a5c8bc427`): Let's see if this watch has fired. It says: "Fires when Anthropic officially announces the restoration of access to Fable 5 and Mythos 5." Wait, Anthropic has *not* officially restored access; in fact, the crackdown has *escalated* because we now know (from the WSJ story published on June 13, 2026) that Amazon's Andy Jassy's talks with the Trump administration triggered a sweeping Commerce Department ban on foreign users, which led to Anthropic shutting down access to Mythos and Fable for *all* users (to ensure compliance and because of foreign-born developers). So access is still shut down, and the situation is even more locked down. So this watch has *not* fired yet. Let's check the other watches: - `Antitrust Scrutiny of Anthropic Silent Safeguards` (id: `019eafd3-288d-7900-9b7b-f29c43234577`): Has an official FTC or EU antitrust investigation been launched? Not yet mentioned in our recent stories. - `Appeals and Rulings on AI Summary Liability in the EU` (id: `019eafd3-1ce9-7376-80d4-c4f5e9580f9d`): Any updates on the Munich Regional Court ruling? No new updates in this cycle's stories. - `Institutional Research Platforms Launch Public MCP Servers` (id: `019ea05f-eaaf-7afb-bb51-55b5f00e17d3`): Has FactSet, PitchBook, or S&P Global launched a *new* public MCP server? Our notes show they did unbundle, but let's see if there's any new event. - `Gemini Deep Research Max Transitions to GA` (id: `019ea05f-d685-7208-a431-55568fba4987`): Has Google DeepMind transitioned Deep Research & Deep Research Max to GA? No new announcement in our current feed. Wait, let's look at how we can update our findings to reflect the latest developments. Let's look at the existing notes and slugs: 1. `claude-fable-5-silent-safeguards-and-pricing-shift`: We should definitely update this note because we have concrete new evidence! - The WSJ revealed on June 13, 2026, that the shutdown was triggered by Amazon CEO Andy Jassy's private conversations with senior Trump administration officials (including Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick). - Amazon researchers found that Fable 5 could be prompted to find security bugs in software programs, which escalated into a wide Commerce Department ban on foreign users. - Anthropic responded by shutting down access to both Fable and Mythos for *all* users, citing compliance and the fact that its foreign-born researchers were effectively blocked from working on the models. - We also have the reaction of Chinese labs like Zhipu (GLM-5.2) and MiniMax/Z.ai releasing frontier open-weight models under the banner of "radical openness" to counter Western restrictions. This is a massive addition to the story of Fable 5, national security, and model access. 2. `market-map-positioning-hey-lefty`: We should update this note to include a highly structured, updated market map of the autonomous and AI-powered research tool landscape in 2026. - We will categorize the market into distinct tiers: - **Tier 1: High-Speed Consumer Search & Synthesis** (e.g., Perplexity Deep Research, offering fast, parallel web RAG). - **Tier 2: Long-Running Multi-Step Reasoning Agents** (e.g., Gemini Deep Research / Max, OpenAI o3-deep-research, conducting deep multi-turn exploration). - **Tier 3: Specialized Academic & Literature Synthesis** (e.g., Elicit, Consensus, Scite, mapping academic papers, clinical trials, and citation networks, contrasting with legacy Google Scholar). - **Tier 4: Structured Financial & Private Market Intelligence** (e.g., Sacra, Tegus, unbundling proprietary private-company data, expert transcripts, and financial models into programmatic APIs and MCP servers). - **Tier 5: Model-Agnostic, Multi-Source Orchestrators (Hey, Lefty's Sweet Spot)**: Acting as the "control plane" that coordinates these various models, open-weight alternatives (like GLM-5.2), and custom data servers (using MCP, ACP, or A2A protocols) without vendor lock-in. - We will also add the **Protocol Interoperability Standards** that are shaping how these research agents interact: - **Model Context Protocol (MCP)** (Anthropic): Vertical integration (model to tools/data). - **Agent Communication Protocol (ACP)** (IBM/Linux Foundation): Horizontal integration (agent-to-agent, peer-to-peer delegation, memory transfer, streaming, async execution). - **Agent-to-Agent (A2A) Protocol** (Google DeepMind/Linux Foundation): Open network standard for cross-cloud agent collaboration. - We will explain how **Hey, Lefty**'s positioning as a model-agnostic, multi-source orchestrator is uniquely resilient to the supply chain risks and geopolitical crackdowns (like the Fable 5 ban) that plague monolithic, single-provider tools. Let's write these two notes! Wait, let's double-check if we need to search or verify anything else. Let's see if there are any other specific questions in the brief: - "What is the market for autonomous or AI research tools? There's gemini deep research, google scholar, perplexity. Sacra is another research platform. What do all of these tools do? What are their features? Their value prop? Their core technology? Their data and where does it come from? Who do they sell to? what is the pricing/business model? Help me build a market map to see where Hey, Lefty fits and we should position it." Yes, the market map note can cover all of these in a beautifully structured, comprehensive, and detailed way, incorporating all the specific competitors (Gemini Deep Research, Google Scholar vs specialized academic tools, Perplexity, Sacra, and OpenAI). Let's draft the notes carefully, ensuring we include concrete quotes and links from our research this cycle. Let's check the quotes we have for `claude-fable-5-silent-safeguards-and-pricing-shift`: We have: - WSJ article: "Amazon CEO’s Talks With U.S. Officials Triggered Crackdown on Anthropic Models" - Quote from WSJ: `"We are talking about a significant escalation in the politicization of AI and centralization of control over advanced computation in this country" — https://www.wsj.com/tech/ai/amazon-ceos-talks-with-u-s-officials-triggered-crackdown-on-anthropic-models-dcc90578?st=Yct6gx&reflink=desktopwebshare_permalink` - Quote from WSJ: `"Researchers at Amazon had used a series of prompts to get Anthropic’s Fable 5 model to provide them with information that could be used to aid cyberattacks and was supposed to be off limits, Jassy told the officials, according to people familiar with the matter."` - Quote from HN user lubujackson: `"Amazon needs Anthropic to bend the knee for their investment to have long term value - the sooner the better." — https://news.ycombinator.com/item?id=48519440` - Quote from Jie Tang (Zhipu founder) on GLM-5.2: `"The path to AGI (Artificial General Intelligence) must never be enclosed by high walls... In the face of external blockades and restrictions, our attitude is one of radical openness." — https://x.com/jietang/status/2065784751345287314` (via `48518684`) Let's check the quotes we have for `market-map-positioning-hey-lefty`: We can cite: - Sacra pricing and documentation: `"Sacra is a research and data platform covering growth-stage and pre-IPO companies. We produce original research — company reports, analyst interviews, revenue models, and curated news... Use our MCP connector to bring Sacra private company data & research into your AI workflows" — https://sacra.com/pricing/` / `https://sacra.com/features/mcp/` / `https://docs.sacra.com/` - IBM Research blog on ACP: `"ACP connects agents to agents; MCP connects agents to their tools and knowledge." — https://research.ibm.com/blog/agent-communication-protocol-ai` - IBM Research blog on ACP: `"Without a common standard, every integration is costly duct tape... Our goal is to build the HTTP of agent communication" — https://research.ibm.com/blog/agent-communication-protocol-ai` - Niklas Heidloff's blog on ACP vs MCP vs A2A: `"Tools are typically stateless; agents can have memory. Model Context Protocol (MCP) defines how to invoke tools. Protocols like Agent Communication Protocol (ACP) define how to invoke agents." — https://heidloff.net/article/mcp-acp-a2a-agent-protocols/` This is highly rigorous, perfectly grounded, and directly answers the user's brief while integrating the latest market events. Let's write the first note: `claude-fable-5-silent-safeguards-and-pricing-shift`. We will update it to reflect the latest details of the Amazon-triggered crackdown and the geopolitical open-weight counter-responses like GLM-5.2. Let's do that.

I will now write the updated finding for `claude-fable-5-silent-safeguards-and-pricing-shift` to incorporate the newly uncovered details of the Amazon-triggered US government crackdown on Anthropic's models, and the geopolitical counter-response from Chinese labs like Zhipu. After that, I will update the market map in `market-map-positioning-hey-lefty` to provide a comprehensive, structured breakdown of the autonomous research tools landscape in 2026, including the emerging protocol standards (MCP, ACP, and A2A). Then, I will conclude the cycle. Let's write the first note.

I will now write the updated finding for `market-map-positioning-hey-lefty` to map the autonomous research tools landscape in 2026, incorporating the core features, technology, and business models of the major competitors, and detailing the emerging interoperability standards (MCP, ACP, A2A). This will explicitly define "Hey, Lefty"'s unique model-agnostic, multi-source orchestration positioning. Let's write the note.

I have updated both notes with comprehensive, structured findings. Let's see if there are any other actions to perform before finishing the cycle. We have some open threads or watches. Let's see if we can resolve any threads. Is there an active thread we can resolve? Let's check `list_threads` output: We had: - `Anthropic Restores Access to Fable 5 and Mythos 5` (id: `019ebf48-b349-73e7-984e-261a5c8bc427`): Watch, not resolved. - `Antitrust Scrutiny of Anthropic Silent Safeguards` (id: `019eafd3-288d-7900-9b7b-f29c43234577`): Watch, not resolved. - `Appeals and Rulings on AI Summary Liability in the EU` (id: `019eafd3-1ce9-7376-80d4-c4f5e9580f9d`): Watch, not resolved. - `Institutional Research Platforms Launch Public MCP Servers` (id: `019ea05f-eaaf-7afb-bb51-55b5f00e17d3`): Watch, not resolved. - `Gemini Deep Research Max Transitions to GA` (id: `019ea05f-d685-7208-a431-55568fba4987`): Watch, not resolved. Wait, are there other open threads of kind `question` or `investigation`? The `list_threads` output returned only these 5 threads, all of which are of kind `watch` (cardinality `one_shot` or `ongoing`). Wait, let's verify if we need to open any new threads. We can record a watch or thread on: - How the Linux Foundation manages A2A and ACP protocol convergence (since both are now under Linux Foundation governance). - When Zhipu's open-weights GLM-5.2 is officially released as open weights (it's currently under the coding plan but not yet fully downloadable as open weights, as noted by some HN comments). Let's register a watch for the full open-weights release of GLM-5.2, or the convergence of ACP and MCP protocols, since they are highly relevant to our positioning. Let's record a watch for: `Zhipu Releases Full Open-Weights for GLM-5.2`. - Type: `tripwire` or `dated_catalyst` (if we don't have a date, `tripwire` or `thesis_breaker`). Let's make it a `tripwire` or `dated_catalyst` (cardinality: `one_shot`). Let's call `record_watch` for this.

This is excellent. Let's also record a watch for the potential convergence or joint-standardization of MCP and ACP (or if Anthropic transfers MCP governance to an open foundation like the Linux Foundation). Let's call `record_watch` for this.

Everything is set up beautifully. I have written the findings and recorded the watches. I will now call `finish_cycle` to complete this research cycle.

I have completed the research cycle successfully. Here is a summary of the findings and updates recorded in this cycle: ### 1. Geopolitical Escalation & Model Access Supply Risks * **Finding Updated:** `claude-fable-5-silent-safeguards-and-pricing-shift` * **Key Discoveries:** * The sudden US government export ban on Anthropic's Claude Fable 5 and Mythos 5 models was triggered by private conversations between **Amazon CEO Andy Jassy** and senior Trump administration officials, including Treasury Secretary Scott Bessent. Jassy warned that Amazon researchers had successfully prompted Fable 5 to expose security bugs in several software programs, raising concerns about automated exploit generation. * In response, Anthropic shut off access to Fable and Mythos for *all* users worldwide to ensure compliance and manage the risk of foreign-born developers working on the models under the ban. * This regulatory blockade triggered an immediate geopolitical counter-offensive from Chinese AI labs. Zhipu AI released **GLM-5.2**, a highly capable open-weight model with a 1M context window. Zhipu's founder, Jie Tang, explicitly framed the release as a stance of "radical openness" against Western model blockades and high walls enclosing frontier intelligence. ### 2. Comprehensive 2026 Autonomous Research Market Map * **Finding Updated:** `market-map-positioning-hey-lefty` * **Key Discoveries:** * **The Stratified Market:** We mapped the autonomous research space into four distinct tiers: * *Tier 1 (High-Speed Consumer RAG):* Perplexity Deep Research (optimized for fast, parallel web synthesis). * *Tier 2 (Long-Running Reasoning Agents):* Gemini Deep Research/Max and OpenAI o3-deep-research (orchestrating multi-step, dynamic research tasks over hours). * *Tier 3 (Academic & Literature Synthesis):* Elicit, Consensus, and Scite (mapping citation networks and systematic reviews over structured academic databases). * *Tier 4 (Structured Financial & Private Market Intelligence):* Sacra and Tegus (unbundling expert transcripts, company profiles, and financial models into APIs and MCP servers). * **Protocol Interoperability Standards:** We analyzed the critical standard protocols shaping the agentic web: * *Model Context Protocol (MCP) [Anthropic]:* Vertical integration (connecting a single model to tools/knowledge). * *Agent Communication Protocol (ACP) [IBM/Linux Foundation]:* Horizontal integration (connecting agents to other agents as peers, supporting streaming, persistent memory across sessions, and asynchronous execution/human-in-the-loop). * *Agent-to-Agent (A2A) Protocol [Google DeepMind/Linux Foundation]:* Open standard for cross-cloud agent collaboration. * **Hey, Lefty's Sweet Spot:** Positioned as the *model-agnostic orchestration layer and control plane* sitting on top of this fragmented ecosystem. In an era of sudden government model bans and silent safeguards, Hey, Lefty's ability to dynamically route research tasks between proprietary models (OpenAI, Gemini), local open-weight rigs, and open-weight alternatives (GLM-5.2) guarantees operational sovereignty and geopolitical resilience. It orchestrates vertical MCP data connections alongside horizontal ACP peer-to-peer delegation, maintaining a dynamic, living workspace with persistent tracking and automated watches. ### 3. New Watches Registered * **Watch 1:** `Zhipu AI Releases Full Open-Weights for GLM-5.2` (monitoring when the weights are publicly downloadable on Hugging Face/GitHub). * **Watch 2:** `Anthropic Transfers MCP Governance to Open Source Foundation` (monitoring the open governance transition or potential convergence of MCP and ACP).