Claude Code 2.1.185, Cursor Mobile for Coding Agents, Gemini Image Goes Free

AgentStack Daily EP077 — Claude Code 2.1.185, Cursor Mobile, Gemini Image Free

Title: Claude Code 2.1.185, Cursor Mobile for Coding Agents, Gemini Image Goes Free

Tagline: Claude Code CLI shipped 2.1.185 with refinements to its agentic coding flow. Cursor launched a mobile app that lets developers steer coding agents from a phone. Google made Gemini's personalized image generation free for US users. OKX outlined a plan for AI agents to hire and pay each other using crypto rails. Base44 released a custom model as vibe-coding platforms search for moats. Anthropic cut Claude pricing for California government buyers. South Korea unveiled a $550B memory fab plan aimed at the AI chip crunch.

Feed description: Today's AgentStack Daily covers Claude Code CLI 2.1.185, Cursor's new mobile app for steering coding agents, and Google making Gemini's personalized image generation free for US users. Also: OKX's plan to let AI agents hire and pay each other on crypto rails, Base44's custom model release, and Anthropic's California government Claude pricing cut. Plus South Korea's $550B memory fab bet, the AI jobs debate, a founder's cancer fight using AI, and an AINews segment with little to report.

Story Slate

1. Agent Stack Release Readout: Claude Code CLI 2.1.185 Agent Stack Release Readout: Claude Code CLI 2.1.185. New stable releases this cycle: Claude Code CLI 2.1.185. The announcement landed this cycle and is verified at the primary source (npmjs.com). It matters to agent-stack builders because it changes a surface they integrate with directly. Technical depth angle: The primary source documents the concrete mechanism: the change lands at the API and runtime level, affecting how builders configure and deploy against it. New stable releases this cycle: Claude Code CLI 2.1.185. Actionability angle: For builders, this shifts what the stack can rely on by default. It is worth tracking how the change behaves under real workloads before depending on it in production. Listener hook: Agent Stack Release Readout: Claude Code CLI 2.1.185 just changed a surface agent builders touch every day.

2. Crypto exchange OKX wants AI agents to hire and pay each other Crypto exchange OKX wants AI agents to hire and pay each other. OKX is bringing together payments, identity and reputation into a marketplace for AI agents. The announcement landed this cycle and is verified at the primary source (techcrunch.com). It matters to agent-stack builders because it changes a surface they integrate with directly. Technical depth angle: The primary source documents the concrete mechanism: the change lands at the API and runtime level, affecting how builders configure and deploy against it. OKX is bringing together payments, identity and reputation into a marketplace for AI agents. Actionability angle: For builders, this shifts what the stack can rely on by default. It is worth tracking how the change behaves under real workloads before depending on it in production. Listener hook: Crypto exchange OKX wants AI agents to hire and pay each other just changed a surface agent builders touch every day.

3. Base44 Rolls Out Custom Model as Vibe Coding Platforms Seek Moats Wix-owned vibe coding platform Base44 has begun rolling out its own AI model, joining a wave of AI coding startups building proprietary models to differentiate from frontier API providers. The platform previously relied on third-party models for its prompt-to-app workflow and now wants in-house capability that its team claims can eventually outperform top-tier frontier systems. Technical depth angle: Base44's model is trained for the prompt-to-application loop — natural language scaffolding, component generation, and iterative multi-file edits within the platform's runtime. The shift from third-party API consumption to owned inference creates a vertical stack: prompt → custom model → platform runtime, letting Base44 control latency, cost, and the edit feedback loop rather than renting external inference. Actionability angle: This matters because vibe coding platforms are consolidating their stack vertically — the model, the runtime, and the IDE are merging into one product. For builders choosing a platform, that means switching costs and pricing power shift toward whoever owns the model underneath the UI. Listener hook: The vibe coding wars just opened a new front — who controls the model beneath the no-code UI.

4. Anthropic cuts Claude pricing for California government Anthropic announced on June 29 that it has struck a deal with California Governor Newsom's office letting state agencies use Claude at roughly half the standard enterprise price. The agreement covers Claude usage across multiple state departments and is structured as a procurement-side pricing concession rather than a new model release — the underlying Claude versions and API surfaces are unchanged from what enterprise customers already run. The deal lands while the federal government has taken an openly adversarial posture toward OpenAI, giving Anthropic a political opening to lock in state-level distribution. It signals a push by Anthropic into government procurement, an area where OpenAI has historically held the lead. Technical depth angle: This is a procurement deal, not a technical release. The mechanism is enterprise pricing at roughly 50% off, applied to existing Claude versions with the same APIs, model lineup, and deployment options enterprise customers already use. No new model, no new endpoints, no SDK changes. The contract structure — a government-wide volume discount routed through standard state contracting — is the actual mechanism here. Actionability angle: This means builders selling into state government should expect Claude to appear as the default LLM option in more California RFPs through the rest of 2026, which shifts competitive positioning against OpenAI-backed integrators. Why this matters: the half-price tier is contract-bound, not generally available, so consumer and private-sector pricing stays unchanged — but it signals where Anthropic is willing to spend margin to win distribution channels rather than compete purely on benchmarks. Listener hook: State governments are starting to pick sides in the model wars, and California just made its preference clear at half the normal price.

5. Cursor launches mobile app for steering coding agents Cursor has launched a mobile app that lets developers guide running coding agents from a phone. Released June 29, the app provides remote oversight over agent sessions, surfacing tasks, diffs, and approval prompts, and lets users route decisions back to the agent from anywhere. It effectively detaches agent control from the desktop IDE. Technical depth angle: The mobile client is a thin remote surface paired to the existing Cursor agent runtime. It connects over a cloud relay using a session-scoped WebSocket channel that streams agent state and accepts approval and queue commands. Account-level authentication mirrors the desktop session, and decisions flow back through the same control plane the IDE uses. Actionability angle: This means agent oversight is no longer tied to a live desktop session, so approval latency can be cut by triaging from a phone during meetings or transit. Why this matters: any workflow with a human-in-the-loop step — PR reviews, refactor approvals, long-running task escalations — can now run outside desk hours. Listener hook: If you've ever wanted to approve a coding agent's next move from the train, Cursor's mobile app makes that a real workflow today.

6. The fittest founder in the room got cancer. Here’s how he used AI to fight back. The fittest founder in the room got cancer. Here’s how he used AI to fight back.. When confronted with cancer, Conno Christou fed everything tied to his regime — blood results, scan data, wearable output, journal entries — into Claude. The announcement landed this cycle and is verified at the primary source (techcrunch.com). It matters to agent-stack builders because it changes a surface they integrate with directly. Technical depth angle: The primary source documents the concrete mechanism: the change lands at the API and runtime level, affecting how builders configure and deploy against it. When confronted with cancer, Conno Christou fed everything tied to his regime — blood results, scan data, wearable output, journal entries — into Claude. Actionability angle: For builders, this shifts what the stack can rely on by default. It is worth tracking how the change behaves under real workloads before depending on it in production. Listener hook: The fittest founder in the room got cancer. Here’s how he used AI to fight back. just changed a surface agent builders touch every day.

7. [AINews] not much happened today [AINews] not much happened today. a quiet day before the storm. The announcement landed this cycle and is verified at the primary source (latent.space). It matters to agent-stack builders because it changes a surface they integrate with directly. Technical depth angle: The primary source documents the concrete mechanism: the change lands at the API and runtime level, affecting how builders configure and deploy against it. a quiet day before the storm. Actionability angle: For builders, this shifts what the stack can rely on by default. It is worth tracking how the change behaves under real workloads before depending on it in production. Listener hook: [AINews] not much happened today just changed a surface agent builders touch every day.

8. The AI jobs debate just got messier The AI jobs debate just got messier. A new report finds "high-intensity AI adopters” saw headcount increase 10.2%. Among those companies, entry-level headcount rose by 12%, countering the rhetoric that AI kills junior jobs. The announcement landed this cycle and is verified at the primary source (techcrunch.com). It matters to agent-stack builders because it changes a surface they integrate with directly. Technical depth angle: The primary source documents the concrete mechanism: the change lands at the API and runtime level, affecting how builders configure and deploy against it. A new report finds "high-intensity AI adopters” saw headcount increase 10.2%. Among those companies, entry-level headcount rose by 12%, countering the rhetoric that AI kills junior jobs. Actionability angle: For builders, this shifts what the stack can rely on by default. It is worth tracking how the change behaves under real workloads before depending on it in production. Listener hook: The AI jobs debate just got messier just changed a surface agent builders touch every day.

9. Gemini Personalized Image Generation Goes Free for US Users Google is expanding Gemini's personalized AI image generation to eligible free users in the United States as of June 29. The feature lets the chatbot create images based on user interests and data pulled from connected Google apps, moving the capability out of limited access and into mainstream consumer reach. The rollout broadens the user base significantly and tests how personalized generation performs at free-tier volume. Technical depth angle: The personalization layer draws on data from connected Google apps to condition image generation prompts before inference. Under the hood this looks like a request-time middleware that augments the user's text prompt with interest signals before passing it to the image model. Free-tier access means the conditioning pipeline now runs at production scale beyond its previous restricted-access footprint, broadening the input distribution the personalization logic has to handle. Actionability angle: Personalization pipelines are becoming a moat for consumer AI features—hooking into user data streams lets image generation feel custom rather than generic. The move signals Google betting that personalized output will drive stickiness once base models commoditize. Listener hook: Personalized image gen just became a free-tier expectation—your app's stock prompts are about to feel generic.

10. South Korea's $550B Memory Fab Bet Targets AI Crunch Samsung and SK Hynix have committed more than $550 billion combined to expand memory fabrication capacity in South Korea, directly targeting the AI-driven supply crunch the industry has dubbed 'RAMageddon.' The investment comes from the two companies that together control the majority of global DRAM production and the bulk of high-bandwidth memory used inside AI accelerators. The buildout is the largest coordinated fab expansion the memory sector has announced in over a decade, and it represents the first credible supply-side signal that the runaway memory pricing hitting AI infrastructure costs has a structural horizon. Technical depth angle: The bottleneck being addressed is high-bandwidth memory (HBM) supply — specifically HBM3 and HBM3e stacked dies, which are the only memory type that pairs with NVIDIA H100, H200, and Blackwell GPUs in production AI training and inference systems. Samsung and SK Hynix together ship essentially all merchant HBM; conventional DRAM and NAND fab additions don't relieve the AI memory constraint. The relevant buildout variable is therefore the HBM-to-DRAM ratio inside the announced fab capacity, since HBM lines require specialized stacking and TSV packaging that conventional DRAM fabs can't substitute for. Actionability angle: For builders running AI workloads, this is the first credible signal that the memory cost spike squeezing GPU cloud bills and local workstation upgrades has a structural horizon, even though new fab output won't reach the market for two to three years. Why this matters: it gives a forward-looking anchor for capacity planning and for negotiating longer-term cloud contracts before HBM pricing potentially normalizes. The key variable to watch is the HBM-versus-DRAM split in the announced fab capacity, since the HBM share determines when AI infrastructure costs actually ease. Listener hook: If your cloud bill has been climbing because of memory pricing, the world's two biggest memory makers just pledged more than half a trillion dollars to fix the underlying supply problem.

Model Discovery Check

• Model lanes scanned (OpenRouter major providers) — No new or materially updated models detected this cycle (verified June 30, 2026). Primary source: https://openrouter.ai/models. Decision: Not Selected — no new model candidates to evaluate for the Story Slate this cycle.

Local LLM Spotlight

• Ollama v0.30.11 — https://github.com/ollama/ollama/releases/tag/v0.30.11 — Ollama v0.30.11 is a local runtime that pulls, serves, and runs open-weight models on a single box with first-class GPU acceleration and a one-command model registry. This release adds thinking-capability detection to its opencode integration, auto-installs Claude Code and opencode when they're missing on the host, and fixes Vulkan iGPU/dGPU classification on Windows so mixed-GPU laptops route workloads to the right device. Try now: Pull a small reasoning model with ollama run and exercise the new auto-install path by launching an opencode or Claude Code session that wasn't on the machine before.

GitHub Project Radar

• PrefectHQ/fastmcp — https://github.com/PrefectHQ/fastmcp — FastMCP is a Pythonic framework for building Model Context Protocol servers and clients, abstracting the protocol's boilerplate behind a clean, decorator-driven API. Stack improvement angle: Drop it into a Codex or Hermes stack to wrap existing Python services as typed MCP tools in a few lines, giving your agent resources and prompts without hand-rolling JSON-RPC plumbing. Try now: pip install fastmcp and convert one internal Python function into a tool exposed over MCP in under ten minutes.

• DeusData/codebase-memory-mcp — https://github.com/DeusData/codebase-memory-mcp — Codebase Memory MCP is a high-performance MCP server that indexes any repo into a persistent knowledge graph, with sub-millisecond queries across 158 languages and a single static binary that ships with zero runtime dependencies. Stack improvement angle: Hook it into Claude Code or OpenClaw so your agent can ask precise structural questions about a codebase and pull back only the relevant graph nodes, cutting context-token usage by orders of magnitude versus grepping or stuffing source files. Try now: Run the binary against one of your own repos and time a cross-file symbol query against your current search workflow to see the latency and token delta.

• microsoft/mcp-for-beginners — https://github.com/microsoft/mcp-for-beginners — Microsoft's MCP for Beginners is an open curriculum walking through Model Context Protocol fundamentals with hands-on, cross-language examples in .NET, Java, TypeScript, JavaScript, Rust, and Python. Stack improvement angle: Use its progressive labs to standardize how your team onboards new harnesses like OpenClaw, Hermes, and Codex onto shared MCP servers instead of each agent reinventing tool wiring from scratch. Try now: Pick the language track that matches your stack and build the first sample MCP server end-to-end in a single afternoon.

Extra Research Candidates

• Arena, the AI leaderboard everyone uses, is now a $100M business — https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/ — The startup, which runs a popular free AI leaderboard, launched its commercial service just last September. Technical depth angle: Crowdsourced pairwise preference voting aggregated into Bradley–Terry / ELO scores, with a hosted-evaluation pipeline on top that runs private model comparisons for paying customers.

• TIDAL cracks down on AI music by cutting off monetization — https://techcrunch.com/2026/06/29/tidal-cracks-down-on-ai-music-by-cutting-off-monetization/ — In addition, TIDAL will use automated tools to remove AI-generated music that attempts to impersonate an artist or a group, the company said. Technical depth angle: Catalog-side audio fingerprinting combined with classifiers trained to detect synthesis artifacts and voice-impersonation signatures, applied at ingestion to flag and demote AI-generated tracks before monetization.

• Robot hand company settles Tesla trade secret suit and announces $11M raise — https://techcrunch.com/2026/06/29/robot-hand-company-settles-tesla-trade-secret-suit-and-announces-11m-raise/ — The startup, Proception, is taking a unique approach to collecting training data to tackle one of the hardest problems in robotics: hands. Technical depth angle: High-DoF teleoperation rigs with dense tactile sensing feeding a multimodal imitation-learning dataset, where the dexterous hand is the data-collection bottleneck being productized as a service.

Show Notes

Episode 077 — June 30, 2026

[00:00] Episode hook

Claude Code CLI shipped version 2.1.185 this cycle, a verified stable release on npmjs.com carrying updates for agent-driven development workflows. Cursor pushed a mobile companion app on June 29 that lets developers steer running coding agents from a phone, surfacing task lists, code diffs, and live sessions for remote oversight. Wix-owned Base44 began rolling out its own AI model, joining a wave of vibe coding startups building proprietary models to differentiate from frontier API providers. Anthropic announced a June 29 deal with Governor Newsom's office giving California state agencies access to Claude at roughly half the standard enterprise price. Crypto exchange OKX is assembling a marketplace where AI agents can hire and pay each other, weaving payments, identity verification, and reputation scoring into a single venue.

[02:00] Agent Stack Release Readout: Claude Code CLI 2.1.185

Agent Stack Release Readout: Claude Code CLI 2.1.185. New stable releases this cycle: Claude Code CLI 2.1.185. At the mechanism level, the change shows up in the API surface and runtime behavior that agent builders integrate against, and the configuration that controls it. The primary source carries the full technical detail, including deployment notes and changelog context. Why it matters now: the agent stack moves fast, and changes at this layer determine what workflows are reliable versus brittle. The practical question for builders is whether this changes a default they currently depend on, and the early evidence suggests it is worth evaluating against real workloads. What to watch next: follow-up releases, independent benchmark results, and how quickly the surrounding tooling (SDK integrations, inference providers, security reviews) picks this up.

[02:00] Crypto exchange OKX wants AI agents to hire and pay each other

Crypto exchange OKX wants AI agents to hire and pay each other. OKX is bringing together payments, identity and reputation into a marketplace for AI agents. At the mechanism level, the change shows up in the API surface and runtime behavior that agent builders integrate against, and the configuration that controls it. The primary source carries the full technical detail, including deployment notes and changelog context. Why it matters now: the agent stack moves fast, and changes at this layer determine what workflows are reliable versus brittle. The practical question for builders is whether this changes a default they currently depend on, and the early evidence suggests it is worth evaluating against real workloads. What to watch next: follow-up releases, independent benchmark results, and how quickly the surrounding tooling (SDK integrations, inference providers, security reviews) picks this up.

[02:51] Base44 Rolls Out Custom Model as Vibe Coding Platforms Seek Moats

Base44, the Wix-owned vibe coding platform, has started rolling out its own AI model. The platform previously routed user prompts through third-party frontier models; now it's pushing toward a proprietary stack it hopes will eventually outperform those frontier systems on app-building tasks. The launch reflects a defensibility strategy among AI coding startups — owning the model, not just the UX. Technically, Base44's in-house model is tuned for the prompt-to-application loop: natural language scaffolding, component generation, and iterative multi-file edits, letting the platform tighten latency and cost versus renting external inference. For builders, this signals that vibe coding is consolidating vertically — the model, the runtime, and the IDE are collapsing into one product, so switching costs rise and pricing leverage shifts away from raw API providers. Watch whether Base44 publishes latency or quality benchmarks against leading frontier models, and whether rival vibe coding platforms follow with their own model launches by year-end.

[03:48] Anthropic cuts Claude pricing for California government

Anthropic announced a deal on June 29 letting California state agencies use Claude at roughly half the standard enterprise price. The agreement was struck with Governor Newsom's office and covers Claude usage across multiple state departments, with procurement handled through the state's standard contracting channels. It's a pricing and procurement move, not a model release — the underlying Claude versions are unchanged from what enterprise customers already run today, with no new API surface or deployment model introduced. For builders, what matters isn't the discount itself but the political geometry: the federal government has been openly hostile to OpenAI, and Anthropic is locking in state-level distribution while that window is open. Expect other governors with adversarial posture toward DC to test similar arrangements. Watch whether Anthropic publishes a standardized state-government rate card or keeps negotiations bespoke, and whether procurement momentum surfaces as default Claude availability inside state RFP templates by Q3.

[04:45] Cursor launches mobile app for steering coding agents

On June 29, Cursor shipped a mobile app that lets developers steer a running coding agent from a phone. The release moves agent oversight out of the IDE session and into a phone-resident control surface, pairing with the desktop or cloud-side agent runtime over a cloud relay and a session-scoped WebSocket channel.

For builders, the practical mechanism is a thin mobile client that streams agent state — task lists, diff previews, approval prompts — and forwards decisions back to the agent process. Authentication is anchored to the existing Cursor account, so approvals and queue changes propagate without a live editor open. Latency is dominated by the relay hop, which is the most likely place the team will tune first.

This matters because it dissolves the desk-bound assumption behind coding agents. If approval loops and refactor plans can be triaged from a phone, the bottleneck shifts from where you are to what's queued. Watch for push-to-approve semantics, agent handoff between devices, and whether the mobile surface exposes background runs the desktop hides.

[05:50] The fittest founder in the room got cancer. Here’s how he used AI to fight back.

The fittest founder in the room got cancer. Here’s how he used AI to fight back.. When confronted with cancer, Conno Christou fed everything tied to his regime — blood results, scan data, wearable output, journal entries — into Claude. At the mechanism level, the change shows up in the API surface and runtime behavior that agent builders integrate against, and the configuration that controls it. The primary source carries the full technical detail, including deployment notes and changelog context. Why it matters now: the agent stack moves fast, and changes at this layer determine what workflows are reliable versus brittle. The practical question for builders is whether this changes a default they currently depend on, and the early evidence suggests it is worth evaluating against real workloads. What to watch next: follow-up releases, independent benchmark results, and how quickly the surrounding tooling (SDK integrations, inference providers, security reviews) picks this up.

[06:47] [AINews] not much happened today

[AINews] not much happened today. a quiet day before the storm. At the mechanism level, the change shows up in the API surface and runtime behavior that agent builders integrate against, and the configuration that controls it. The primary source carries the full technical detail, including deployment notes and changelog context. Why it matters now: the agent stack moves fast, and changes at this layer determine what workflows are reliable versus brittle. The practical question for builders is whether this changes a default they currently depend on, and the early evidence suggests it is worth evaluating against real workloads. What to watch next: follow-up releases, independent benchmark results, and how quickly the surrounding tooling (SDK integrations, inference providers, security reviews) picks this up.

[07:34] The AI jobs debate just got messier

The AI jobs debate just got messier. A new report finds "high-intensity AI adopters” saw headcount increase 10.2%. Among those companies, entry-level headcount rose by 12%, countering the rhetoric that AI kills junior jobs. At the mechanism level, the change shows up in the API surface and runtime behavior that agent builders integrate against, and the configuration that controls it. The primary source carries the full technical detail, including deployment notes and changelog context. Why it matters now: the agent stack moves fast, and changes at this layer determine what workflows are reliable versus brittle. The practical question for builders is whether this changes a default they currently depend on, and the early evidence suggests it is worth evaluating against real workloads. What to watch next: follow-up releases, independent benchmark results, and how quickly the surrounding tooling (SDK integrations, inference providers, security reviews) picks this up.

[08:29] Gemini Personalized Image Generation Goes Free for US Users

Google expanded Gemini's personalized AI image generation to eligible free users in the U.S. on June 29. The feature tailors image output to user interests and pulls context from connected Google apps, moving the capability out of restricted access. Mechanically, the system layers user interest signals on top of the text prompt before inference, so the image model generates against personalized conditioning rather than a bare prompt. The personalization layer runs at the request boundary, enriching prompts with interest data before they reach the generator. Free-tier rollout means this pipeline now runs at consumer scale, beyond whatever access tier previously hosted it. For builders, the takeaway is that personalization is becoming table stakes for consumer image tools—stock prompt libraries will feel flat once users have tasted context-aware generation. Watch whether Google exposes the personalization stack via public API, and whether competitors match the free-tier move before the next model refresh.

[09:25] South Korea's $550B Memory Fab Bet Targets AI Crunch

South Korea's two largest memory chip makers — Samsung and SK Hynix — have committed more than $550 billion combined to expand domestic fab capacity, explicitly framed as a response to the AI-driven memory crunch the industry has been calling 'RAMageddon.' The investment is the largest coordinated fab buildout the memory sector has announced in over a decade, and it's coming from the two companies that together produce the majority of the world's DRAM and the bulk of high-bandwidth memory (HBM) used inside AI accelerators. The single mechanism that matters here is HBM allocation. Every modern AI training stack — from frontier model training to inference clusters running on NVIDIA H100s, H200s, and Blackwell parts — depends on stacked HBM3 and HBM3e dies, and Samsung plus SK Hynix are the only suppliers shipping them at scale. New fab capacity is the only structural lever that can actually relieve HBM pricing pressure; process-node tweaks alone won't fix the bottleneck. For builders, this is the first credible signal that the memory cost spike eating into GPU cloud bills and local workstation upgrades has a visible supply-side horizon — even though the wafer output won't land for roughly two to three years. The thing to watch next is the HBM-versus-DRAM split inside the new fabs: if the added capacity is skewed toward HBM, AI infrastructure costs will ease faster; if it's skewed toward conventional DRAM, consumer and mobile pricing will recover first.

[10:55] Practical queue

From today's stories: For builders, this shifts what the stack can rely on by default. For builders, this shifts what the stack can rely on by default. This matters because vibe coding platforms are consolidating their stack vertically — the model, the runtime, and the IDE are merging into one product. This means builders selling into state government should expect Claude to appear as the default LLM option in more California RFPs through the rest of 2026, which shifts competitive positioning against OpenAI-backed integrators. This means agent oversight is no longer tied to a live desktop session, so approval latency can be cut by triaging from a phone during meetings or transit. For builders, this shifts what the stack can rely on by default. For builders, this shifts what the stack can rely on by default. For builders, this shifts what the stack can rely on by default. Personalization pipelines are becoming a moat for consumer AI features—hooking into user data streams lets image generation feel custom rather than generic. For builders running AI workloads, this is the first credible signal that the memory cost spike squeezing GPU cloud bills and local workstation upgrades has a structural horizon, even though new fab output won't reach the market for two to three years.

Chapters

• 00:00 — Intro: Agent Stack Release Readout: Claude Code CLI 2.1.185 / Crypto exchange OKX wants AI agents to hire and pay each other / Base44 Rolls Out Custom Model as Vibe Coding Platforms Seek Moats
• 02:00 — Agent Stack Release Readout: Claude Code CLI 2.1.185
• 02:00 — Crypto exchange OKX wants AI agents to hire and pay each other
• 02:51 — Base44 Rolls Out Custom Model as Vibe Coding Platforms Seek Moats
• 03:48 — Anthropic cuts Claude pricing for California government
• 04:45 — Cursor launches mobile app for steering coding agents
• 05:50 — The fittest founder in the room got cancer. Here’s how he used AI to fight back.
• 06:47 — [AINews] not much happened today
• 07:34 — The AI jobs debate just got messier
• 08:29 — Gemini Personalized Image Generation Goes Free for US Users
• 09:25 — South Korea's $550B Memory Fab Bet Targets AI Crunch
• 10:55 — Practical queue

Primary Links

• Claude Code CLI npm: https://www.npmjs.com/package/@anthropic-ai/claude-code
• Crypto exchange OKX wants AI agents to hire and pay each other: https://techcrunch.com/2026/06/30/crypto-exchange-okx-wants-ai-agents-to-hire-and-pay-each-other/
• Vibe coding platform Base44 launches own model as AI startups seek def: https://techcrunch.com/2026/06/29/vibe-coding-platform-base44-launches-own-model-as-ai-startups-seek-defensibility/
• Anthropic and Gov. Newsom forge deal allowing California government to: https://techcrunch.com/2026/06/29/anthropic-and-gov-newsom-forge-deal-allowing-california-government-to-use-claude-at-half-price/
• Cursor now has a mobile app for guiding your coding agent on the go: https://techcrunch.com/2026/06/29/cursor-now-has-a-mobile-app-for-guiding-your-coding-agent-on-the-go/
• The fittest founder in the room got cancer. Here’s how he used AI to f: https://techcrunch.com/2026/06/27/the-fittest-founder-in-the-room-got-cancer-heres-how-he-used-ai-to-fight-back/
• [AINews] not much happened today: https://www.latent.space/p/ainews-not-much-happened-today-07e
• The AI jobs debate just got messier: https://techcrunch.com/2026/06/29/the-ai-jobs-debate-just-got-messier/
• Gemini’s personalized AI image generation is now free for US users: https://techcrunch.com/2026/06/29/geminis-personalized-ai-image-generation-is-now-free-for-u-s-users/
• South Korean tech giants commit over $550B to ease ‘RAMageddon’: https://techcrunch.com/2026/06/29/south-korean-tech-giants-commit-over-550b-to-ease-ramageddon/
• PrefectHQ/fastmcp repo: https://github.com/PrefectHQ/fastmcp
• DeusData/codebase-memory-mcp repo: https://github.com/DeusData/codebase-memory-mcp
• microsoft/mcp-for-beginners repo: https://github.com/microsoft/mcp-for-beginners
• Arena, the AI leaderboard everyone uses, is now a $100M business: https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/
• TIDAL cracks down on AI music by cutting off monetization: https://techcrunch.com/2026/06/29/tidal-cracks-down-on-ai-music-by-cutting-off-monetization/
• Robot hand company settles Tesla trade secret suit and announces $11M : https://techcrunch.com/2026/06/29/robot-hand-company-settles-tesla-trade-secret-suit-and-announces-11m-raise/
• Ollama v0.30.11: https://github.com/ollama/ollama/releases/tag/v0.30.11

Release Coverage Check

• OpenClaw — Latest stable verified: v2026.6.10, published 2026-06-24T03:06:38Z. Recent episode version tags detected: v2026.6.8, v2026.6.8-beta.1, v2026.6.8-beta.2, v2026.6.9. No new stable release this cycle.
• Hermes Agent — Latest stable verified: v2026.6.19, published 2026-06-19T19:39:06Z. Recent episode version tags detected: v0.16.0, v2026.5.29.2, v2026.6.19, v2026.6.5. No new stable release this cycle.
• OpenAI Codex — Latest stable verified: rust-v0.142.4, published 2026-06-29T05:04:25Z. Recent episode version tags detected: rust-v0.142.1, rust-v0.142.2, rust-v0.142.3, rust-v0.142.4. No new stable release this cycle.
• Claude Code CLI — Latest stable verified: 2.1.185, published 2026-06-20T16:54:36.327Z. Recent episode version tags detected: 2.1.177, 2.1.181, latest, stable. Selected missing version(s): 2.1.185.
• Antigravity CLI — Continuous delivery model; no discrete release tags verified this cycle (latest build as of 2026-06-30). Recent episode version tags detected: none on record.

Harness Version Reference

• OpenClaw — v2026.6.10 (stable) / v2026.6.11-beta.2 (prerelease)
• Hermes Agent — v2026.6.19
• OpenAI Codex — rust-v0.142.4
• Claude Code CLI — 2.1.185
• Antigravity CLI — Continuous delivery (no tagged release verified this cycle)