Anthropic turns coding sessions into live, shareable web pages — so a whole team can watch an agent's work update in real time, instead of waiting for a status report.
On June 18, Anthropic rolled out Artifacts inside Claude Code, letting a terminal session publish its own progress as a living web page. Built from the full context of a coding session — the codebase, connected tools, and the conversation itself — it can render PR walkthroughs, dashboards, or release checklists that keep updating as the work continues.
The most common early use case, by Anthropic's own account, is debugging: an engineer kicks off an incident investigation, and instead of writing a stand-up update, shares a link to a page that's already tracking the timeline, the suspect commits, and the error rate — updating in place as the session digs deeper.
Every publish creates a new version at the same URL, so teammates can roll back to an earlier state, and a gallery keeps every artifact ever generated in one place. It's in beta for Team and Enterprise plans, reachable from the CLI and desktop app — a quiet admission that AI-written work increasingly needs a human-readable surface, not just a terminal log.
Full story · ClaudeA self-improving memory layer for Perplexity's Computer agent learns from its own past work — not just from your preferences.
Most AI memory is about you — your preferences, your tone, your recurring requests. Perplexity's new Brain, launched June 18 for its Computer agent, is built around something different: the work itself. It logs what an agent tried, what worked, what failed, and what corrections a person made along the way.
Overnight, that log gets synthesized into a kind of personal wiki the agent reads before its next task, so a project doesn't start from zero every morning. Perplexity says early internal testing showed real movement on tasks the system had already seen before, alongside a drop in the cost of pulling up historical context.
It's a research preview limited to Max and Enterprise Max subscribers for now, and the company has been careful to frame it as something narrower than general intelligence — an agent getting better at its own job, one correction at a time.
Full story · DecryptA new --temporary flag in Wrangler gives coding agents a 60-minute scoped Cloudflare account — no human, no OAuth flow, no copy-pasted tokens.
An AI agent that can write a working app in seconds still hits a very human wall the moment it tries to ship it: a sign-up page. Cloudflare's answer, rolled out this week, lets any agent deploy a Worker without an account, a password, or a browser-based login.
The deployment goes live on a workers.dev URL immediately and stays claimable for 60 minutes — long enough for an agent to iterate, redeploy, and hand a human a link to make it permanent. Skip the claim, and the whole thing quietly expires.
It's a small mechanic with a large implication: Cloudflare is betting that a growing share of its traffic will arrive from background agents with no person watching the screen, and that the platforms which remove friction first are the ones agents will default to.
Full story · CloudflareOpenAI's ChatGPT for Excel and Google Sheets is now generally available across every plan, turning plain-language prompts into live formulas and models.
ChatGPT for Excel and Google Sheets graduated out of beta this spring and is now generally available across every paid and free tier, running on GPT-5.5 inside a sidebar that lives directly in the spreadsheet rather than a separate chat window.
The pitch is one most spreadsheet users will recognize instantly: describe what you want — a budget tracker, a cleaned-up dataset, a formula that's been silently wrong for months — and the assistant builds or fixes it in place, asking permission before it touches a cell and linking its answers back to the exact ranges it used.
It still can't touch pivot tables, macros, or Power Query, and OpenAI is upfront that outputs need a human check before anyone trusts a board deck to it. But for the unglamorous bulk of spreadsheet work, it's now a default rather than a beta.
Full story · OpenAIGenesis AI's first general-purpose robot, Eno, skips legs and a face entirely — built on wheels, with hands that match human dexterity instead.
Every robotics company chasing humanoid form right now is implicitly betting that the future looks like us. Genesis AI's first general-purpose robot, Eno, makes the opposite bet: a wheeled base, a tower of articulated panels that folds down to roughly the size of a suitcase, and hands built to match human dexterity rather than a human face.
Industrial customers reportedly asked for the wheels themselves — flat factory and lab floors don't need legs, and legs mostly add cost and failure points. Eno's brain, a foundation model called GENE, is meant to let it hold a goal across an extended task rather than execute a single pre-scripted motion.
Genesis plans deployments to manufacturing, logistics, and lab customers before the end of 2026, including a strategic rollout partnership announced alongside the unveiling.
Full story · Fast CompanyHonor's gimbal-armed Robot Phone resurfaced in high-resolution press images this month, ahead of a Q3 2026 launch.
Honor's "Robot Phone" resurfaced in high-resolution press images this week, courtesy of a Qualcomm anniversary event — a reminder that the concept first teased at Mobile World Congress in March is still very much on track for a global launch later this year.
The whole pitch hinges on one mechanism: a robotic-arm gimbal that retracts flush into the phone's body, then pops out and unfolds with a single tap to physically reposition the camera — tracking a subject, following a face through a video call, or swinging into a cinematic angle mid-shot.
It's an unusually literal answer to "embodied AI" from a smartphone maker, built on Honor's existing foldable-hinge expertise rather than anything borrowed from humanoid robotics.
Full story · TechNodeA $29.99-a-month Premier tier gives Waymo's heaviest riders priority pickups and cash back — the robotaxi business growing up into a subscription.
Waymo just did something no robotaxi operator had tried before: it launched a paid membership. Waymo Premier costs $29.99 a month, invite-only for now in San Francisco, Los Angeles, and Phoenix — the company's three longest-running markets.
Members get priority vehicle matching, up to five free cancellations a month, and 10% of every fare back as Waymo Cash, with early access promised whenever the service opens in a new city. It's a familiar loyalty playbook, borrowed straight from Uber One and Lyft Pink, just applied for the first time to a car with no driver in it.
Waymo says the price and perks were shaped directly by rider feedback. The timing isn't subtle either — the company is mid-expansion toward 20-plus cities and an international launch in London later this year, and a subscriber base gives it a built-in audience the moment each new market opens.
Full story · TechCrunchGemini-powered AI Overviews now summarize long threads and surface action items directly inside Gmail.
Gmail's slow absorption into "the Gemini era" kept moving this year, and the most useful piece for anyone drowning in client threads is the simplest one: AI Overviews now read an entire email conversation and hand back a few sentences of what actually matters — decisions made, questions still open, who owes whom a reply.
Point it at a label instead of a single thread, and it does the same triage across everything filed under it — ask what's outstanding under "Clients," and it surfaces the people waiting on you, the documents that were promised, and the open questions, without opening a single email by hand.
It's rolling out as a free, default feature for everyone, with deeper natural-language Q&A over your whole inbox gated behind a Google AI Pro or Ultra subscription.
Full story · Google