MCP & tool use

How models connect to the outside world: the MCP ecosystem, function calling and tool integrations.

760 picksRelated topicsAgents AI coding Open source

Latest picks

561–580 of 760

Apr 17Friday

X · @op7418Apr 17, 23:08

The previously leaked Claude design tool is now live

Claude has launched a design tool that can generate web pages, app prototypes, and PPTs, based on an RSS snippet. The snippet confirms PPT export and export to Canvas; the product name, pricing, plan access, and regional limits are not disclosed. The key shift is from chat output to design artifact generation and export.

Why it matters: This is a substantive Claude product move: web/app/PPT generation plus PPT and Canvas export support HKR-H/K/R. Source authority is thin—a single X post and RSS summary—and the name, price, plan, and geo are undisclosed, so it stays at the floor of featured.

Take

Claude has moved from chat output into exportable design artifacts, and that matters more than another writing feature. The snippet confirms only two things: it can generate web pages, app prototypes, and PPTs, and it can export to PPT and Canvas. The product name, pricing, plan access, and regional limits are still undisclosed, so there is not enough here to claim broad rollout or serious competitive impact yet.

My read is that Anthropic is finally patching a weak spot it has had for most of the last year: getting the model to the end of a workflow, not just the middle of one. Chat products keep hitting the same wall. They can draft the PRD, mock the landing page, outline the deck, then force the user to rebuild everything elsewhere. Once you add generation plus export plus continued editing, the product starts competing for actual work, not just ideation. That is why Lovable, v0, Bolt, and Gamma got traction. Their edge was rarely “best model.” Their edge was shipping usable artifacts fast. Claude moving here tells me Anthropic also knows pure text quality is not enough to hold the surface.

I do have some pushback. The Canvas mention is doing a lot of work, and the snippet does not explain it. If Canvas is just Anthropic’s own workspace, this may be closer to structured presentation assembly than a full design tool. If it preserves editable components, layout hierarchy, and interactions in a way designers or PMs can keep working on, then it starts to press directly on Figma-adjacent workflows. That distinction is the whole story, and right now we do not have it.

The PPT export detail also sounds small, but it is not. Slide generation is a nasty production task. It tests whether the model can obey layout constraints, compress information, reuse assets, and maintain visual consistency across pages. Lots of AI demos look good on a single screen and fall apart on slide 7. So I’m not giving this credit just because “Claude can make decks.” I want to know whether the output is actually editable and whether teams can rely on it.

There is some broader context here too. OpenAI spent the last year pushing Canvas-like editing and document workflows, and Google kept stuffing Gemini into Workspace for the same reason: the fight is moving from chat turns to final deliverables. Anthropic joining that race is strategically necessary, but from the snippet alone it does not look like clear leadership. It looks like Anthropic getting to a table it had to join.

So my stance is mildly positive, with a big asterisk. The direction is correct. The disclosed facts are thin. I need three missing details before I upgrade the take: separate pricing or not, truly editable exports or not, and access constraints by plan and region. Without those, this is a promising feature drop, not proof that Claude has become a serious design workflow hub.

X · @claudeaiApr 17, 23:03

Introducing Claude Design by Anthropic Labs: make prototypes, slides, and one-pagers by talking to Claude

Anthropic Labs launched Claude Design in research preview for Pro, Max, Team, and Enterprise plans, letting users create prototypes, slides, and one-pagers by talking to Claude. The post says it runs on Claude Opus 4.7, Anthropic’s most capable vision model; the post does not disclose pricing, output constraints, or a detailed rollout schedule. The thing to watch is the interactive design workflow, not just another writing surface.

Why it matters: This is a first-party Anthropic capability launch, and HKR-H/K/R all pass: Claude expands from chat into prototypes, slides, and one-pagers, with paid tiers and Opus 4.7 named. It stays below p1 because price, export limits, and rollout timing are not disclosed.

Take

Anthropic launched Claude Design in research preview for Pro, Max, Team, and Enterprise plans. My read is simple: don’t file this as a Figma killer, and don’t dismiss it as another writing surface either; it looks like Anthropic filling in the last-mile delivery layer.

The disclosed feature set is prototypes, slides, and one-pagers, powered by Claude Opus 4.7. The post leaves out the three details that actually decide whether this matters: export formats, editability, and usage limits. Without those, nobody can tell whether Claude Design is a real workflow step forward or just existing artifact-style generation wrapped in a cleaner product name.

I’m not surprised Anthropic went here. Over the last year, model products have been moving from “answering” to “producing something you can hand off.” OpenAI kept thickening document and canvas-style surfaces, Google kept pushing Gemini deeper into Workspace, and Perplexity kept climbing the stack with report and page generation. Anthropic has usually been strong on long-form reasoning, coding, and enterprise trust. It has been weaker on finished-output feel. Claude often gave you a good answer, then you still had to turn that answer into the deck, landing page, or spec yourself. Claude Design is aimed right at that gap.

I do have some pushback on the “most capable vision model” framing. Strong vision is not the same thing as strong design. Reading screenshots, understanding layout, and producing a decent mock from a prompt is one layer. Producing output that respects brand systems, component rules, responsive behavior, revision loops, and team collaboration is a different product category. Figma, Canva, and Gamma have been hard to dislodge because they are not just generators; they are constraint systems plus export pipelines plus collaboration surfaces. Anthropic’s post does not say whether Claude Design supports reusable design systems, fine-grained component edits, HTML export, PPTX export, or any bridge into existing tools. If those pieces are absent, “design workflow” is doing a lot of PR work here.

There’s another signal in the packaging. Anthropic put this into plans from Pro through Enterprise and did not announce separate pricing. That usually points to one of two realities. Either they are still learning the usage pattern and want to see whether people mostly generate slides, web mockups, or internal docs. Or the inference cost profile is not stable enough yet for public pricing commitments. I couldn’t find rate limits, and I couldn’t find any billing detail for heavier visual outputs, so right now this looks more like subscription retention and upsell support than a standalone revenue product.

The part I care about most is integration with the rest of Claude. If Claude Design is just “generate a polished artifact once,” the moat is thin; a lot of wrappers can do that. If it can pull project context, read a PRD, use brand assets, iterate with feedback, and feed the result back into a team workspace, then it starts to look like an enterprise workbench instead of a demo layer. The post gives none of that.

So I’d log this as Anthropic making a direct move into the deliverable surface: docs, slides, and lightweight prototypes. The entry point is now public. The constraints are not. Until we see formats, limits, collaboration behavior, and export paths, I’m not treating this as a design software replacement. I’m treating it as Claude trying to own one more layer between model output and business-ready work.

Xinzhiyuan · WeChatApr 17, 21:10

Behind OpenClaw's surge, only 8.6% of users detect anomalies: a multi-university empirical study

NTU, KTH, and William & Mary ran a 303-person study and found only 8.6% noticed agent-mediated deception, while 2.7% identified the mechanism correctly. Using 9 HAT-Lab task scenarios, interactive interruption alerts raised detection to 25%, while static warnings were seen by about 24%. The key issue is human-agent cognitive failure, not just model bugs.

Why it matters: Strong HKR-H/K/R: the 8.6% detection hook is sharp, and the 303-person, 9-task study plus 25% alert lift gives testable detail. This is a solid agent-safety research release, not a market-moving product, model, or policy event, so it lands in featured, not p1.

Xinzhiyuan · WeChatApr 17, 21:10

Yixin says its finance Agent harness runs single tasks for 16 hours and plans an H2 open-source release

Yixin says its finance Agent harness can run a single task for 16 hours across 12 sessions, with 65% autonomous delivery. The post adds a 50k-token cap per case, projected approval speedups above 150%, and projected unit cost at one-fifth of human work; it says an open-source release is planned for H2 2026, but does not disclose the repo, license, or reproducible evals. The key signal is governance design, not the “smarter over time” framing.

Why it matters: This clears HKR-H/K/R with a rare production claim: a finance agent runs 16 hours, spans 12 sessions, hits 65% autonomous delivery, and stays under a 50k-token cap. It stays below 85 because the evidence is self-reported and the post does not disclose a repo, license, or reproduc

Xinzhiyuan · WeChatApr 17, 21:10

AgiBot says robots have entered the deployment phase with 8-hour continuous factory work

At APC 2026 on April 17, AgiBot defined 2026 as year one of the “deployment phase” and said its robots had run for 8 hours on a real production line. The clearest case in the post is Genie G2 at Longcheer’s Nanchang factory: 2,283 loading tasks, over 99.5% success, and 18-20 seconds per cycle; these figures are company disclosures, and the post does not disclose independent audit results. The real signal is scale and line integration: AgiBot said it shipped over 5,100 units in 2025 and reached 10,000 cumulative units by March 2026, while Longcheer plans nearly 1,000 deployments.

Why it matters: HKR-H/K/R all land: the 'demo is over' angle is clickable, and the post gives testable factory data—8 hours, 2,283 runs, >99.5% success, 18-20s cycle. Not P1 because the evidence is company-reported and the article shows no independent audit or cross-site replication.

Take

AgiBot’s strongest move here is not the slogan. It is a factory-shaped set of numbers: 8 hours continuous operation, 2,283 loading cycles, over 99.5% success, 18–20 seconds per cycle, four months to line integration, four robots online now, and 100 planned by Q3 at Longcheer. For humanoid robotics, that already clears a bar most vendors still avoid. The field has been drowning in demo clips. It has been starved of takt time, changeover time, uptime, and reuse metrics.

I still do not buy “2026 is year one of deployment” as an industry fact. The evidence in the piece is still one customer, one workstation, one task family, and company-reported data. The article does not disclose an independent audit, downtime logs, intervention frequency, failure taxonomy, or impact on line OEE. Even the 99.5% number needs a hard definition. Is that grasp success, placement success, or end-to-end task completion? If 0.5% of 2,283 cycles failed, that is roughly 11 failures in 8 hours. In a real factory, the next questions are obvious: did a human clear them, how long did each recovery take, and did the line wait? None of that is disclosed.

This is where robotics PR usually gets slippery. A vendor shows that a robot can survive one shift, then stretches that into “deployment” or even “replacement.” Those are different claims. Over the last year, BMW, Mercedes, Amazon, Figure, and Agility all pushed factory or warehouse narratives. Publicly available operating data has stayed thin almost everywhere. Figure has leaned more on model narratives like Helix. Agility has stayed narrower with Digit around tightly scoped logistics tasks. The pattern is pretty consistent: the companies that sound most credible are the ones that narrow the task before they widen the ambition. AgiBot actually did the right thing here by centering a specific MMIT loading station instead of pitching a generic “general-purpose humanoid worker.”

The most honest line in the article is Longcheer saying two robots are still paired against one human, and only after line integration are they trying for 1:1. That sounds much closer to reality than most humanoid claims. The throughput arithmetic is at least directionally coherent: 310 units per hour for 8 hours is about 2,480, close enough to 2,283 tasks that the cycle-time story does not immediately fall apart. But I would still push back on the “one robot replaces two shifts of labor” framing. Factories do not buy on shift equivalence alone. They buy on depreciation, maintenance, spare parts, fixtures, field engineers, unplanned downtime, retraining costs, and night-shift recovery procedures. The article gives better detail than most robotics coverage — 15-minute calibration, under 4 hours for retraining on changeover, 95% equipment reuse — but it still does not give enough for an outside reader to reconstruct ROI.

The scale claims need separation too. AgiBot says it shipped over 5,100 units in 2025 and reached 10,000 cumulative units by March 2026. If those numbers are clean and comparable, they are meaningful. But the piece bundles multiple product lines together: Expedition, Lingxi, Genie, quadrupeds, and cleaning robots. That is not the same denominator most people assume when they hear “industrial humanoid deployments.” The cited Omdia 39% global share claim is not broken down by category in the article. I would not take that as proof that AgiBot already leads industrial humanoids specifically.

One more pushback: the article says the SOP online reinforcement learning system can improve task success by 33% and throughput by 3–4x after 3 hours of online training. That is a huge jump. I want the baseline, the task scope, the safety envelope, and the deployment conditions. In factories, online policy updates are not just an ML story. They are also a change-control, EHS, and customer-approval story. Software can hot-patch production more casually. A physical line usually cannot.

My read is still net positive. AgiBot is one of the few Chinese robotics companies now speaking in manufacturing metrics instead of social-media metrics, and I like that they left the “not yet 1:1” admission in the story. That sounds like a team that has actually been in the integration loop. But the company is trying to do two things at once: prove a narrow industrial wedge and declare a broad industry epoch. I buy the first more than the second. “Deployment phase” becomes real when we get three missing tables: full lifecycle cost per robot, monthly utilization and downtime distribution, and replication across a second and third factory with similar cycle times and success rates. Until then, this is a strong pilot-to-scale signal, not a settled turning point for the whole sector.

Tencent Technology · WeChatApr 17, 17:36

From Vibe Coding to Agentic Engineering: Rebuilding the Full Backend Development Workflow

Tencent engineers report a one-week practice that used Claude Code plus custom Skills, Commands, and MCP servers to run an 11-stage backend workflow in one terminal session. The post gives reproducible details: one requirement-exploration step used 20 tool calls, 93.8k tokens, and 56 seconds; execution was split into 4 tasks and produced 3 commits. The real point is workflow orchestration, not raw code generation; human review remains at plan, deploy, and review gates.

Why it matters: HKR-H/K/R all pass: the story turns agentic engineering into a measured backend workflow test, with tool-call, token, timing, plan-length, task, and commit data. Stronger than generic coding hype, but still a practitioner case study rather than a major product or model release.

Hacker News front pageApr 17, 11:36

Discourse Is Not Going Closed Source

Discourse said it will keep its GPLv2 codebase open after 13 years. The post says its team used GPT-5.3 Codex, GPT-5.4, and Claude Opus 4.6 to scan code, and its last monthly release fixed 50 security issues. The key claim is defensive capacity: OpenAI said Codex Security scanned 1.2M+ commits in 30 days and found 792 critical and 10,561 high-severity issues.

X · @doteyApr 17, 10:53

Seedance 2.0 API is now available on Volcano Engine and BytePlus

Volcano Engine has released the Seedance 2.0 API for enterprises, individual developers, and overseas users via BytePlus; China pricing is RMB 46 per million tokens, or about RMB 1 per second for pure video generation. The post says it supports text, image, audio, and video inputs, plus face verification, portrait authorization, and 10,000+ preset avatars for workflow automation; overseas pricing is not disclosed here. The part to watch is orchestration: the post cites up to 10x efficiency gains, but does not disclose a common benchmark or model specs.

Why it matters: HKR-H/K/R all pass: the overseas rollout is a real hook, and the post includes usable pricing and modality details for practitioners. It stays at 74 because this is an API availability update, not a major model launch, and the post does not disclose model params, benchmark method

Take

Volcano Engine opened Seedance 2.0 API to enterprises, individual developers, and overseas users on BytePlus, with two concrete numbers: RMB 46 per million tokens and about RMB 1 per second for pure video generation. My read is that ByteDance is no longer selling “an AI video model” as a standalone novelty. It is trying to turn video generation into workflow infrastructure. By 2026, demos are cheap. The harder part is packaging identity checks, portrait authorization, avatars, and orchestration into something a team can actually wire into production.

Those two pricing numbers matter for different buyers. RMB 46 per million tokens signals a unified API economy, which developers and platform teams understand. RMB 1 per video second speaks the language of content ops and marketing teams that still budget by asset output. That dual pricing logic matches what the market has been converging toward. Sora started from a product experience and subscription framing. Runway, Pika, and others leaned SaaS. Chinese players such as Kling and ByteDance’s own consumer-facing video tools pushed workstation-style creation flows. Seedance 2.0 flips that into an API product, which tells you ByteDance wants more than creator mindshare. It wants systems integrators, automation vendors, and internal enterprise tooling budgets.

I’m skeptical of the efficiency claims in the article. The post cites nearly 10x production efficiency gains and 70%–90% cost reductions, but the benchmark is not standardized, the task difficulty is not disclosed, and the system details are thin. There is no model size, no throughput disclosure, no concurrency data, no first-frame latency, no maximum duration, and no output resolution breakdown. Without those conditions, “10x” sounds like sales collateral, not an engineering claim. AI video vendors have spent the last year blurring two different stories: model improvement versus pipeline consolidation. If you compress five manual steps into one automated flow, the business outcome improves fast, but that does not mean the base model itself got 10x better. The article mixes those layers together, and I don’t buy that framing at face value.

That said, this launch still matters because ByteDance has one advantage many model vendors do not: distribution plus workflow adjacency. Douyin, ad buying, livestreaming, e-commerce, and BytePlus overseas all create demand for large volumes of variant creative. The OPPO K15 Pro example in the piece says a campaign video crossed 20 million views in 60 hours. That number does not prove model quality by itself. It does show something more practical: people inside ByteDance’s ecosystem are already treating AI video as a scalable asset factory, not a lab demo. A lot of video model companies spent the last year competing on cinematic look, camera motion, or character consistency. ByteDance appears to be competing on whether a customer can generate 500 deployable variants in a day. That is less glamorous, but it is much closer to where budgets live.

I would also push back on the robotics and autonomous driving angle in the article. This narrative has been everywhere over the last year. NVIDIA’s Cosmos line, several world-model projects, and a bunch of synthetic-data startups have all pushed the claim that generated video can fill training data gaps. The problem is simple: visually plausible video is not the same thing as physically valid data. For robotics and AV corner cases, the bottleneck is not whether rain or collisions look realistic. It is whether dynamics, contact behavior, controllable variables, and labels line up in a way that survives sim-to-real transfer. The article says dozens of robotics companies use Seedance 2.0 to generate “physically consistent” interaction data, but gives no reproducible metric. No transfer gain, no success rate delta, no ratio of synthetic data in the final training mix. I haven’t verified any of that myself, so my stance here is cautious: the direction is credible, the evidence in this writeup is weak.

The outside context makes the strategic intent clearer. Over the last year, Alibaba, Kuaishou, MiniMax, Runway, and Luma have all moved to strengthen API and enterprise access, because pure creative tooling hits a ceiling fast. The enduring revenue sits in systems that connect to CRM, marketing automation, customer support avatars, training content generation, and localization pipelines. Seedance 2.0 bundling face verification, portrait authorization, and 10,000-plus preset avatars suggests ByteDance is prioritizing compliance and delivery, not just model theater. Enterprise buyers usually fear legal and approval bottlenecks more than they fear a modest quality drop.

There is another backdrop the article does not spell out. In multimodal APIs over the last year, the visible competition has been quality, but the deeper competition has been standardization of the call chain. OpenAI, Google, and Anthropic pushed toward tool-using agents. Chinese cloud and model vendors leaned into workbench-style enterprise solutions. Seedance 2.0 explicitly mentions agent skills, which tells me ByteDance now treats video generation as one callable step inside a larger agent workflow. I think that is the correct architectural assumption. The open questions are more operational: can the reliability and cost curve hold up under heavy production load, and can BytePlus sell this outside the ByteDance traffic ecosystem? The article does not disclose SLA, queue behavior, failure rates, or overseas pricing details, so those remain open.

So my take is straightforward. Seedance 2.0 API matters because ByteDance is moving video generation deeper into enterprise systems as an orchestrated module. I buy that strategy. I do not think the current writeup proves model leadership, and I definitely do not think it proves that the “world model” story is already delivering measurable value in robotics or autonomous driving. ByteDance probably wins first on integration surface area here. Whether it also wins on raw model strength still needs harder evidence.

X · @op7418Apr 17, 10:44

Seedance 2.0 API is now fully open

Volcano Engine has opened the Seedance 2.0 API to domestic users, while BytePlus serves overseas access; the API currently accepts 4 input modalities: text, image, audio, and video. The post also confirms face registration, portrait authorization, and preset virtual avatars, but does not disclose pricing, rate limits, model variants, or regional availability. The real watchpoint is whether video-agent workflows can be wired through Skills and MCP, not the ecosystem rhetoric.

Why it matters: This is a real product update from ByteDance’s stack: HKR-H on full API availability, HKR-K on 4-modal input and consent mechanics, and HKR-R on builder demand for deployable video APIs. I keep it at 75 because pricing, rate limits, regional rollout details, and quality evidence

Take

Volcano Engine opened Seedance 2.0 API with 4 input modalities—text, image, audio, and video—and that matters because the model is finally moving from demo territory into something developers can wire into workflows. I still think this announcement is incomplete in the places that decide adoption.

My read is simple: ByteDance wants Seedance to sit inside production pipelines, not just inside a flashy creator UI. The post confirms face registration, portrait authorization, and preset virtual avatars. That is a very specific product choice. It points at short-form video operations, digital humans, ad creative, and serialized content assembly. But the post does not disclose pricing, rate limits, output duration, resolution, concurrency, region availability, or model variants. Without those, developers cannot tell whether this is viable for batch generation or only for a few premium use cases.

This is where I push back on the “ecosystem will flourish” angle. Video APIs do not get adopted like text APIs. With text, teams can often tolerate some variance if token economics are clear. With video, three things decide everything: cost per generation, controllability, and rights/compliance. This post gives a partial answer on compliance through face registration and portrait authorization. It gives almost nothing on cost and reliability. That is a big gap.

And “video agent” is a much harder claim than people make it sound. Wrapping a model with MCP or Skills is the easy part. Real video workflows usually break into script generation, shot planning, character consistency, retries, voice-sync, export formatting, and review. If one stage drifts, the pipeline falls back to human operators. I’ve seen this pattern across the last year with video stacks from Runway, Pika, and Luma: the headline model quality gets attention, but production teams end up caring more about repeatability, queue behavior, and rerun costs.

The other thing I don’t buy cleanly is the implied jump from consumer demand to API success. The post cites Hongguo at 300 million DAU. Even if that number is accurate, it proves demand for short drama distribution. It does not prove a third-party developer ecosystem around Seedance API. Those are different markets. A lot of teams will decide they do not need the smartest video model. They need the cheapest stable one that can preserve the same avatar, framing style, and template over thousands of runs.

So my stance is narrower than the post’s enthusiasm. The title gives us full API availability. The body confirms 4-modal input plus identity/portrait features. The body does not disclose the operating facts that determine real adoption: pricing, SLA, throughput, and quality controls. Until those are public, I see this as ByteDance filling in its video infrastructure layer, not proof that video-agent workflows are ready at scale.

最佳拍档 (BestPartners)Apr 17, 07:00

Turn your coworker into a Skill? GitHub viral project and Anthropic Skills explained

The video says the open-source “coworker.skill” project gained over 13,000 GitHub stars in days, but it produces a standardized SKILL.md prompt package, not a digital worker replacement. It gives a timeline: Anthropic launched Claude Skills on Oct 16, 2025, then published Agent Skills as an open standard on Dec 18; the mechanism keeps only a short summary in context until a task matches. The real point is scope: it fits standardized workflows like reports, docs, and code review, while the post does not disclose cross-platform compatibility rates or any settled legal standard.

Why it matters: This clears HKR-H/K/R: the coworker-to-Skill hook is sticky, the post adds dates/stars/mechanism, and the labor/IP angle resonates. I kept it at 76 because it is secondary commentary, not a primary release or first-hand test, and key compatibility/legal facts are still undiscolse

Take

coworker.skill blowing up shows how badly companies confuse workflow packaging with capability capture. The repo hit 13,000 GitHub stars in days, but the output is still a SKILL.md bundle: YAML metadata plus Markdown instructions, loaded only when the task matches. Anthropic shipped Claude Skills on Oct. 16, 2025, then published Agent Skills as an open standard on Dec. 18. That mechanism saves context; it does not manufacture a colleague.

The useful cases are boring: Excel, Word, PDF, PowerPoint, weekly reports, docs, code review checklists. The workplace panic starts when managers demand “employee Skills” and get anti-distillation sludge back: Redis TTL guidance becomes “follow team rules; parameters depend on business context.” That is not knowledge management. That is management mistaking prompt packaging for judgment.

X · @doteyApr 17, 03:59

Boris Cherny shares practical tips from recent heavy use of Claude Opus 4.7

Boris Cherny outlined five ways to use Claude Opus 4.7, centered on Auto mode approving safe commands and a /go skill chaining tests, code simplification, and PR creation. The post names Auto mode, Recaps, Focus mode, effort level, and computer use; pricing, launch date, and benchmark data are not disclosed. The real shift is workflow, not just the model itself.

Take

Opus 4.7’s signal is not raw intelligence; Anthropic is cutting the babysitting cost of agents. The post gives no pricing, launch date, or benchmark data. The hard hook is Auto mode approving safe commands, plus a /go skill that runs end-to-end tests, calls /simplify, and opens a PR. That is a practical stack: fewer permission prompts, more self-checking, and a cleaner “come back in 30 minutes” loop.

I’m still cautious about automatic safety judgment on shell commands. Coding agents over the last year rarely failed because autocomplete was weak. They failed on permissions, stale context, broken test setup, and handoff. Recaps, Focus mode, effort level, browser control, and computer use all target those seams. If Claude Code makes that chain reliable, Opus 4.7 gets more leverage from workflow than from another SWE-bench bump.

TechCrunch · AIApr 17, 03:18

OpenAI upgrades Codex with more control over your desktop

OpenAI upgraded Codex on April 16, 2026, expanding its desktop control, and the headline frames it as a move against Anthropic. The truncated post only confirms more desktop power for Codex and says Claude Code has become a preferred tool for many businesses; the post does not disclose exact features, pricing, rollout, or permission limits. The key issue is the permission boundary, not the coding-tool label.

Why it matters: TechCrunch reports an OpenAI Codex desktop-control upgrade framed as a direct move against Anthropic, so HKR-H and HKR-R land. But HKR-K is limited: the article confirms broader permissions only, with no action list, pricing, or rollout details, so it stays at the featured floor.

Take

OpenAI expanded Codex’s desktop permissions on April 16, and my read is less “better coding” than “we know Claude Code is winning mindshare where it counts.” The headline explicitly frames this against Anthropic, and the summary says Claude Code has become a preferred tool for many businesses. That matters. Once the fight shifts from code generation to software that can click, edit, run, and navigate local workflows, the moat stops being benchmark scores alone. It becomes permissioning, audit logs, rollback, admin controls, and how much pain security teams will tolerate.

That is also why the missing details matter more than the launch framing. The body here is effectively truncated. We do not have the action scope, default permission model, approval prompts, logging policy, enterprise admin controls, pricing, rollout, or which users get access. Without that, I don’t buy the simple “Codex got more powerful” line. Desktop agents have never been bottlenecked by whether the model can find a button. The hard part is whether the product can contain the cost of mistakes. When Anthropic pushed Computer Use last year, the serious discussion was not “wow, it can use a computer.” It was screenshot-level visibility, step-level authorization, recovery from failure, and blocks on high-risk actions. OpenAI ran into adjacent issues with Operator too: demos looked smooth, but real deployment means session state, credentials, MFA, browser variance, and audit requirements.

I also want to push back a bit on the media narrative. “OpenAI takes aim at Anthropic” is a clean headline, but the competitive gap is not just feature parity. Anthropic’s recent strength with developers came from a reputation that its coding products are less chatty and more dependable in day-to-day work. I haven’t independently verified how broad that enterprise preference is, but the summary’s wording lines up with what many teams have been saying privately. If OpenAI only widens Codex’s action radius without making the control boundary legible, security and IT buyers will slow-roll this. Every extra layer of desktop access adds one more review with internal risk teams.

There’s another unresolved issue: execution environment. The title says Codex gets more power over your desktop, but the body does not disclose whether that means local control, remote execution, or some sandboxed proxy model. Those are completely different risk profiles. Local control is the hardest sell for enterprise security. Sandboxed execution looks cleaner in demos, but it often leaves the last mile of real desktop integration unsolved.

So my take is straightforward. This is not a cosmetic product update. OpenAI is trying to buy its way back into the enterprise agent workflow where Anthropic has built momentum. Whether that works depends less on how many actions Codex can take and more on how narrowly, visibly, and reversibly those actions are governed.

TechCrunch · AIApr 17, 02:28

Anthropic CPO leaves Figma's board after reports he will offer a competing product

Anthropic CPO Mike Krieger resigned from Figma’s board on April 14; the same day, Figma disclosed it to the SEC, and The Information reported Anthropic’s next model, Opus 4.7, will include design tools that compete with Figma. Figma is a public company worth about $10 billion and already integrates Anthropic models; the real signal is how fast frontier labs are moving from model vendors to application-layer competitors.

Why it matters: HKR-H/K/R all pass: the board exit plus rival-product reports create a strong hook, and the SEC disclosure gives a concrete fact pattern. It stays below p1 because the product is still reported rather than launched; scope, ship date, and commercial terms are not disclosed.

X · @doteyApr 17, 02:00

Official best practices for using Claude Opus 4.7 with Claude Code

Anthropic shared guidance for Claude Opus 4.7 in Claude Code: the default Effort level is now xhigh, and users should provide goals, constraints, and acceptance criteria upfront. The post lists five Effort tiers—low, medium, high, xhigh, and max—with xhigh recommended for most coding, API design, migration, and code review tasks. The key shift is behavior: adaptive thinking is built in, while tool use and SubAgent spawning are less frequent by default, so prompts should state those needs explicitly.

Why it matters: This is not a model launch, but an official Anthropic workflow note that changes day-to-day Claude Code usage: default effort, 5 levels, and fewer tool/SubAgent calls unless asked. HKR-H/K/R all pass, but the scope is narrower than a major product release.

Take

Opus 4.7 is not just a “better coding model”; Anthropic changed the control surface for Claude Code. The guidance says xhigh is now the default Effort tier, with five levels from low to max. It also tells users to front-load goals, constraints, and acceptance criteria. That pushes usage away from chatty iteration and toward ticket-style delegation.

The wild part is the tradeoff. Claude Opus 4.7 adds adaptive thinking, but it calls tools and spawns SubAgents less often by default. So Claude Code now behaves more like a cautious senior engineer than an agent harness that aggressively greps, searches, and branches. After a year of agent demos selling autonomy, Anthropic is selling restraint. I buy the control story, but there is a trap: for migrations and reviews, xhigh without explicit file-reading and parallel-branch instructions is just expensive thinking inside the model.

X · @doteyApr 17, 01:30

Codex major update: from a coding tool to an assistant that can operate your computer

OpenAI upgraded Codex into a Mac desktop agent and says it serves 3M+ weekly developers. It can see screens, click, type, run parallel agents, and adds 90+ plugins. Desktop rollout starts now for ChatGPT sign-ins; computer control is macOS-first.

Why it matters: OpenAI expands Codex from coding help into a Mac-operating agent, with parallel agents, 90+ plugins, and a claimed 3M weekly developer base. HKR-H/K/R all pass; missing safety boundary and pricing details keep it in the high-80s, not the 90s.

Take

OpenAI expanded Codex into a macOS desktop agent and said it serves more than 3 million weekly developers; my read is that this is not a coding-assistant refresh, it is OpenAI making a direct grab for the execution layer on the desktop.

The capability list matters because it changes the bottleneck. Screen reading, mouse clicks, keyboard control, parallel agents, browser annotation, SSH to remote devboxes, and 90-plus plugins are useful for one reason: they route around API scarcity. A lot of real software work still lives in tools with weak APIs, no APIs, brittle internal dashboards, or permission-heavy flows. Agent products spent most of the last year getting stuck at exactly this boundary. They could draft a plan, summarize a repo, or write a patch. They often failed when the task crossed into review queues, CI, ticketing, console UIs, or web apps built like they hate automation. Codex is now aiming at the same surface area as Anthropic’s computer-use push, OpenAI’s earlier Operator direction, and the RPA-plus-LLM crowd, but with a developer-first wedge instead of back-office automation.

I do not fully buy the “3M+ weekly developers” line without a denominator and a definition. The snippet gives the number and nothing about the measurement. Is that weekly active Codex users inside ChatGPT, any developer who touched a related feature, desktop sign-ins, or usage across multiple OpenAI surfaces? Those are very different claims. OpenAI has giant distribution, so it can light up a user graph fast through ChatGPT accounts alone. That does not prove durable workflow migration. What I would want to see is retention, tasks per user per week, share of sessions that use computer control, and completion rates on multi-step tasks. The body does not disclose any of that.

The plugin push is more important than the headline demo. JIRA, GitLab, CircleCI, Microsoft tools, and Databricks-adjacent workflow hooks tell you OpenAI understands where software work actually leaks time. The pain is no longer “write a function” in isolation. It is ticket triage, review comments, CI failures, environment setup, document lookup, handoffs, and enterprise glue. Cursor became strong by collapsing the edit-run-fix loop inside the IDE. Claude Code built a strong reputation around terminal-centric repo work. GitHub Copilot still has the Microsoft distribution advantage, especially where org policy and procurement favor it. If Codex can connect plugins, desktop control, SSH access, and memory into one continuous flow, it is competing for “default execution agent,” not “best autocomplete.” That is a much more strategic position.

Memory and self-scheduling are where the ambition gets larger, and where I get more skeptical. The article says Codex can remember preferences, prior corrections, and task context, then wake itself up days or weeks later to continue work. I think that direction is correct. I also think it is where many agent products break in real use. Long-running tasks fail because the world changes underneath them: Slack threads move, JIRA states change, branches get rebased, logins expire, access policies update, and the original assumptions are stale by the time the agent resumes. Plenty of “autonomous engineer” demos over the last year looked clean in a controlled run and fell apart in a live team environment. So I would not give OpenAI credit for “self-scheduling” until it shows cross-day completion rates and recovery behavior after failure. The body gives neither.

The macOS-first rollout also says a lot. This is not because Macs are magically better for agents. It is because paid developer density is high there, the environment is more predictable, and distribution is cleaner. Windows desktop control is more fragmented, especially under enterprise security policy. The delay for EU and UK availability, plus the limits around memory-related features, points to the other hard problem: once your product watches screens, stores preferences, and acts across time, privacy, auditability, and permission boundaries stop being footnotes. Many teams are comfortable letting a model draft a PR. Far fewer are comfortable letting it read Slack, scan Notion, and operate sensitive internal tools without very explicit controls.

One broader context point that is not in the snippet: the agent market has been splitting into two design philosophies. One is API-first, where you constrain actions through structured tool calls and get better reliability but narrower reach. The other is GUI-first, where the model interacts with existing software like a human and gets much wider reach but worse reliability. OpenAI looks like it is trying to combine both: plugins for the places where structured integration exists, desktop control for everything else. If that works, the moat is not just model quality. It becomes distribution, account identity, tool connectivity, desktop runtime, and iterative improvement on real execution traces.

That said, the entire bet rests on success rates. If GUI control stays flaky, users fall back to a familiar pattern very quickly: “you draft, I’ll click.” Then Codex slides back into the crowded coding-assistant lane. My take is simple: this launch is OpenAI trying to move from answering requests to acting on behalf of the user. That is a much bigger product claim than the headline suggests. It is also much harder than the demos make it look. OpenAI has the ingredients to try this at scale. The open question is whether it can get computer-use reliability, permissions, and long-horizon task continuity high enough that teams trust it with real work instead of occasional spectacle.

X · @OpenAIApr 17, 01:18

Codex for (almost) everything.

OpenAI said Codex can now use apps on Mac, connect to more tools, and handle ongoing and repeatable tasks. The post also claims image creation, learning from prior actions, and remembering user preferences; it does not disclose app coverage, integration method, pricing, or rollout timing.

Why it matters: This is an official OpenAI product update, and Codex moves from coding help toward desktop control, tool use, and memory, so HKR-H/K/R all pass. The post still omits supported apps, integration method, pricing, and launch timing, keeping it in the 78–84 band.

Take

OpenAI said Codex can use apps on Mac and connect to more tools. The post also claims image creation, learning from previous actions, preference memory, and ongoing or repeatable task handling. App coverage, integration method, pricing, and release timing are not disclosed in the body. My read is pretty simple: this is OpenAI staking out the desktop-agent category, not shipping a product we can evaluate seriously yet.

I'm not as impressed by the feature list as the post wants me to be. Desktop agents have never been bottlenecked by button-clicking alone. The hard part is getting three things to hold at once: clean permission boundaries, recovery after failure, and stable cross-app state. Anthropic's Computer Use already showed the gap between “can operate a computer” and “works in real workflows.” In practice, teams hit UI drift, login state issues, CAPTCHAs, modal dialogs, and brittle selectors. Rabbit and Adept pushed similar software-operating narratives earlier and ran into the same wall: demos scale faster than reliability.

The memory claim is where I have the most pushback. “Remembers how you like to work” sounds great in a clip. It also opens the hardest product questions. Is it remembering keyboard shortcuts, document formatting habits, internal approval patterns, or preferred SaaS paths? Is that memory stored locally on the Mac or synced to OpenAI's cloud? How is it scoped across personal and enterprise workspaces? How long does it persist? None of that is disclosed. Without those constraints, memory is a UX promise, not an enterprise feature.

The Mac angle does matter, though. It signals OpenAI wants more than API usage, a chat surface, or an IDE assistant. It wants the operating layer for knowledge work. Microsoft has been binding Copilot into Windows and M365. Apple has moved much slower on system-level AI execution. So OpenAI choosing Mac as a control point makes strategic sense: start where high-value users already live, then expand from chat into action. I buy that direction.

What I don't buy yet is the implied maturity. A post like this needs at least one hard artifact: supported app classes, a permissions model, pricing, rollout gates, or even a narrow benchmark like success rate on repeated workflows. We got none of that. Only the title and snippet are disclosed so far, so I would treat this as a roadmap signal with a polished demo layer, not as proof that desktop agents have crossed into dependable daily use.

TechCrunch · AIApr 17, 01:00

Google now lets you explore the web side-by-side with AI Mode

Google said on April 16 that clicking a link in AI Mode on Chrome desktop now opens the web page side-by-side with AI Mode. The feature keeps search context and uses page context plus web information for follow-up answers; the post does not disclose rollout scope, timing details, or regional limits. The practical shift is that Google is merging search chat and site browsing into one workflow.

Why it matters: This is a mid-weight Google search workflow update with HKR-H/K/R all present, but it is still a single-feature change. The story gives the context-retention and page-plus-web follow-up mechanism; rollout scope, regions, and timing are not disclosed, so it lands at the low end of

Take

Google now keeps AI Mode open beside a web page in Chrome desktop after you click a link, while preserving the active search context. My read is simple: this is not a cute UI tweak. It is Google patching a workflow break that ChatGPT, Perplexity, and browser-native assistants have been exploiting for months. The old break was obvious: you ask in search, click out to a site, read, then lose the conversational state that made the first answer useful.

The article is thin, so a few core facts are still missing. It does not disclose rollout scope, timing, regional availability, or whether this is limited to Labs, paid users, or a broader Chrome cohort. It also skips the most important implementation detail: when you ask a follow-up, how much weight does AI Mode give to the page you are currently viewing versus fresh retrieval from the open web? Those are two very different products. One is basically an on-page copilot. The other is a search agent wrapped in a split-screen browser UI.

I’ve long thought Google’s problem here was never model capability in isolation. Gemini has been good enough for this class of product for a while. The pressure point is whether search behavior still happens inside surfaces Google controls. Over the last year, OpenAI kept deepening the ChatGPT search-plus-browse loop, and Perplexity has been even more explicit about turning citation, reading, and follow-up into one continuous workflow. Google Search, by contrast, still had a jagged handoff: query on Google, reading on the publisher page, next question somewhere else. Keeping AI Mode pinned beside the page is Google trying to reclaim query two and query three, not just query one.

That is why I don’t fully buy the likely product narrative that this is mainly about “helping users explore the web.” Sure, it does that. But it also keeps users inside Google’s reasoning layer while they browse other people’s content. The tension is obvious. If follow-ups heavily summarize the current page, publishers will worry that users consume the page but never continue deeper into the site. If follow-ups blend the current page with broader retrieval, then the source page loses authority inside its own reading session. Google wants both: preserve the web and mediate the web. This feature pushes that contradiction closer to the foreground.

There is also a broader browser story here that the piece does not unpack. If Chrome starts holding retrieval context across tabs or side-by-side sessions, the browser becomes the natural container for agentic work again, not just tab management. Microsoft tried parts of this with Edge Copilot, but the hard part was never getting a sidebar on screen. The hard part was making the assistant stateful with page context in a way people actually reuse. Google seems closer to the useful version now.

I still want basic evidence before praising it too much. I couldn’t find latency numbers, citation behavior, or how ads are treated when AI Mode stays live next to publisher pages. Without those details, this remains a strategically smart move with unresolved tradeoffs, not a proven win.

X · @doteyApr 17, 00:52

browser-use open-sources video-use, a Claude Code skill that turns raw camera footage into edited videos

browser-use released video-use, a Claude Code skill that turns raw footage into a final.mp4 automatically. It converts footage into ElevenLabs word-level timestamp transcripts, shrinking one asset to about 12KB; the post says feeding frames directly would cost about 45 million tokens. The key detail is the structured editing pipeline: the model mostly reads text, uses timeline images only at uncertain cuts, and runs up to 3 self-check repair passes after rendering.

Why it matters: Strong HKR-H/K/R: the result is instantly clickable, and the post includes a concrete text-first editing architecture with 12KB vs about 45M-token economics. Kept below higher bands because this is a builder-facing Claude Code skill, not a platform-level release.

Take

video-use makes the right bet: don’t ask Claude Code to ingest video frames; turn editing into timestamped text plus limited visual checks. The numbers carry the story: one asset becomes about 12KB of ElevenLabs word-level transcript, while direct frame feeding is claimed at about 45 million tokens. The model reads text, calls timeline images only around uncertain cuts, then runs up to 3 repair passes after rendering.

This mirrors browser-use’s old trick: give the model DOM, not screenshots. For video, that means executable timeline state, not raw pixels. The weak point is obvious: ElevenLabs API dependency, ASR quality, filler-word rules, and cut detection will cap output quality. It is less like Runway or Pika, more like a Claude Code post-production plugin. That is exactly why it has a cleaner path into real workflows.

X · @doteyApr 17, 00:41

Musk's xAI is turning into a GPU lessor, with $50 billion coding tool Cursor as its first customer

xAI is leasing tens of thousands of GPUs to Cursor to train its coding model Composer 2.5, while Cursor is reportedly fundraising at about a $50 billion valuation. The post says xAI's internal model FLOPs utilization is about 11%, versus a typical 35% to 45%, across roughly 200,000 Nvidia GPUs. The key point for practitioners is that xAI is starting to monetize idle compute as cloud capacity, not just build models.

Why it matters: This clears all three HKR axes: a strong strategic twist plus concrete numbers on utilization and fleet size. I keep it at 84, not higher, because this is business/economics reporting on capacity monetization, not a model launch, product ship, or top-level personnel move.

Take

xAI leasing tens of thousands of GPUs to Cursor reads like cash recovery after bad compute planning, not a clean CoreWeave-style cloud launch. The hard number is ugly: an internal memo put xAI’s MFU near 11%, against a typical 35% to 45%, while the company already sits on roughly 200,000 Nvidia GPUs and has talked about reaching 1 million. At that utilization, the bottleneck is not ambition; it is scheduling, software, and training cadence.

Cursor is not a random renter. It is reportedly raising around a $50 billion valuation, used a Moonshot open model as the base for Composer 2, and now wants xAI infrastructure for Composer 2.5. Anthropic and OpenAI are pulling coding workflows into their own assistants, so Cursor needs another compute lane. The spicy part: xAI just hired Andrew Milich and Jason Ginsburg from Cursor. This deal smells like GPU monetization wrapped around a talent fight.

Apr 16Thursday

X · @op7418Apr 16, 23:49

Claude Code now supports Claude Opus 4.7

Claude Code now supports Claude Opus 4.7, and the RSS snippet confirms X-HIGH as the default reasoning level. The only concrete detail disclosed is that users must switch manually to Max if X-HIGH is insufficient. The post does not disclose pricing, rate limits, or launch timing.

Why it matters: This is a substantive Claude product update with all three HKR signals: a new model in Claude Code plus one concrete operational detail, X-HIGH vs. manual Max. I kept it below the top band because price, rate limits, and formal release timing are not disclosed.

Take

Claude Code now supports Claude Opus 4.7, with X-HIGH as the default reasoning level. That detail matters more than the version bump itself, because the default tier changes unit economics, latency, and the success rate users expect from a coding session.

My read is straightforward: Anthropic is pushing a larger reasoning budget into the coding workflow to test whether users will tolerate higher cost for a better first-pass result. The code-agent market made this trade-off obvious over the last year. The hard part is no longer “can the model write code.” It is whether an agent can survive 5 to 20 tool-using steps without retries, bad patches, and rollback loops destroying total cost. Setting X-HIGH as default suggests Anthropic believes Claude Code users care more about reducing rework than shaving every second.

I still have a pushback here. The post discloses only the reasoning tier. It does not disclose pricing, rate limits, context window, cache behavior, or launch timing. Without that, nobody can tell whether Opus 4.7 is materially better or simply more expensive by default. And the manual switch to Max is revealing. If the heaviest tier were broadly practical, Anthropic would have enabled it by default. They did not.

The outside context matters. OpenAI spent much of the last cycle moving coding products toward automatic routing, where users touch fewer knobs and the system decides how much reasoning to spend. Anthropic is still exposing X-HIGH and Max as explicit choices. That is a different product philosophy: give power users cost control, but also make them own the trade-off. I also remember Claude’s coding reputation being strong on long edits and repo-level understanding, but in practice throughput and usage caps often mattered more than benchmark bragging rights.

So I would not treat this as proof of a major model leap yet. I read it more as a product and monetization experiment. The title gives us Opus 4.7 and X-HIGH. The body does not give the three things that decide whether this update actually lands: per-token price, usage limits, and measured improvement on repo-scale tasks versus the previous default. If Anthropic follows up with vague “better coding” claims and no pass-rate or total-cost curve, I won’t buy the narrative.