an arXiv preprint from Stanford, NYU, MIT and Princeton researchers found a useful warning sign for everyday AI use: people often think AI saves more time than it does on simple tasks like arithmetic, spelling, recall and short rewrites.

across 3 preregistered studies with 2,691 people, participants expected AI to save 55.7 seconds on average, but the measured saving was 7.5 seconds. the bigger finding is behavioral: after using AI on just two tasks, people became more likely to use it again, even when doing the task themselves was faster. the risk is not that AI makes people incapable overnight, but that it can quietly blur the sense of when your own brain is already the quicker tool.

in short: how often are we using AI because it is faster, versus because it makes the work feel lighter?

🔗 https://arxiv.org/abs/2605.22687

Google Research is backing a UC San Diego experiment to turn retired Pixel phones into low-carbon cloud compute. Jeff Dean's framing makes the bet clearer: people replace phones every ~4 years, leaving hundreds of millions of still-usable devices whose manufacturing footprint already exists.

the plan is to remove the batteries and other components, power the boards directly, replace Android's userspace with a general Linux distro, and run 25-50-device clusters under Kubernetes. the first deployment is a planned 2,000-phone UCSD cluster for classes and research, expected in fall 2026.

mobile-phone level compute clusters lol, super cool

🔗 https://research.google/blog/a-low-carbon-computing-platform-from-your-retired-phones/ / https://x.com/JeffDean/status/2065649717573505188

Notra launched image generation for teams that want marketing visuals from the product work they already shipped.

instead of starting from a blank prompt, it uses shipped work, brand style, and real components, then exports into Paper or Figma. useful direction: the same workflow that writes launch posts is now making the visuals around them too.

think visual work for your PRs to show what was done!

🔗 https://www.producthunt.com/products/notra-3 / https://x.com/usenotra/status/2066415660268458204

Cartesia is pushing voice agents as a full stack now: Sonic-3.5 for text-to-speech and Ink-2 for streaming speech-to-text, packaged together for agents that need to listen and respond in real time.

they’re handling: low-latency voice, alphanumeric handling, and turn detection in one API. Ink-2 has the cleaner third-party benchmark story, while Sonic-3.5 is still a strong signal that voice AI is moving from demo voiceovers toward actual phone-agent infrastructure.

excited to give these new models a try!

🔗 https://x.com/krandiash/status/2066559212533190917

Factory 2.0 is not just a new coding-agent demo. the full pitch is a software factory: an org-level feedback loop that turns bugs, customer feedback, requirements, reviews, security, release, and monitoring into agent work that keeps improving.

the useful detail is the maturity curve: simple tasks use Droids or skills, recurring workflows become automations, long-running/local work uses Droid Computers, and bigger jobs become multi-agent Missions. the bet is that engineers will spend more time designing the factory, governance, and context layer around agents, not only writing code directly.

🔗 https://factory.ai/news/software-factory / https://x.com/FactoryAI/status/2066588050617249904

Greptile is adding TREX (TEST, RUN, EXECUTE) a public beta that runs a pull request branch instead of only reading the diff. it spins up a sandbox, mocks api calls, clicks through the UI, writes/runs tests, and attaches evidence like logs, screenshots, traces, or video back to the review.

the important shift is code review agents moving from static comments to runtime checks. Greptile says its own evals caught 20% more bugs with TREX, and it is free for users until the end of June before becoming $2 per run.

🔗 https://www.greptile.com/blog/trex

Salesforce agreed to acquire Fin, formerly Intercom, for about $3.6B. Salesforce is a CRM platform: software that businesses use to manage customer data, sales, marketing, and support. Intercom is the helpdesk/live-chat product many businesses use for website support. Fin is Intercom's AI customer agent, built to answer and resolve customer issues across chat, email, phone, Slack, and other channels.

that's why this matters: this is not just one SaaS company buying another. Intercom renamed itself around Fin only weeks ago, and Salesforce is buying the AI agent layer that can plug into customer data and take support actions. the bet is that customer service shifts from inboxes and tickets to agents that resolve work directly.

🔗 https://www.salesforce.com/news/press-releases/2026/06/15/salesforce-signs-definitive-agreement-to-acquire-fin/
🔗 https://www.intercom.com/blog/today-intercom-becomes-fin/

Playful is trying to make event design feel like something you can do from your phone, not a desktop tool. the thread shows inspiration search, an AI canvas, community remixes, and small edits like QR codes, stickers, fonts, styles, and uploaded assets.

the promo is that they will make the next 888 event covers for free if people repost and tag @playfuldesignco with a request. the more interesting bit is the interface shift: mobile design apps are starting to look less like template pickers and more like prompt, remix, and edit flows.

🔗 https://x.com/barofclo/status/2066547035063595110

anyone who knows me irl knows I’ve been a Cursor fan since 2023, so this one means a lot, especially now that I get to be a Cursor ambassador! (so i get to share joys of using cursor, as well as codex!!)

today’s Cursor Compile announcements point in one direction:

Cursor doesn’t just want to be the editor. Their mission is to be the best way to create software.

1. Cursor Mobile is in TestFlight!
2. Origin looks more like a GitHub/GitLab competitor for agent-heavy workflows, and
3. A much larger in-house model was previewed, reportedly 1.5T+ parameters (Opus and GPT sized) and trained from scratch on 100K+ GPUs with SpaceX.

the bigger idea: Cursor is trying to co-develop the product and the model together. not just people editing pixels on a screen, but giving developers a way to shape deeper parts of the workflow, from model behavior to version control, review, and mobile agents.

🔗 https://cursor.com
🔗 https://cursor.com/origin
🔗 https://testflight.apple.com/join/rPtcW8D2
🔗 https://x.com/SherryYanJiang/status/2066937715581743342
🔗 https://x.com/swyx/status/2066928345246470204
🔗 https://x.com/NickADobos/status/2066927660291498484
🔗 https://x.com/NickADobos/status/2066929277757800833
🔗 https://x.com/RayFernando1337/status/2066928294814237136

Exa launched Exa Agent, a new API for long-running web research, list-building, and enrichment tasks. instead of just returning search results, it can split work across subagents, use Exa's search/content stack, and return text, structured JSON, citations, and a cost breakdown.

the headline claim is price: Exa says its own benchmarks get frontier-style results at less than half the cost of GPT 5.5 and Opus. the more practical bit is that it is available now, with fixed effort tiers from $0.012 to $1 per request, so deep research starts to look more like a priced API call than a bespoke agent workflow.

🔗 https://exa.ai/blog/exa-agent

Framer 3.0 is here, and it turns the website builder into more of an agent workspace. agents can design on the canvas, update CMS content, fix responsiveness, improve SEO, and make code components, while branching lets teams review the changes before they touch production.

the notable part is the external agent path: Codex, Claude Code, Cursor, Gemini CLI, and similar tools can connect into Framer projects. that moves ai site work from one-off page generation toward ongoing site maintenance and team workflows.

🔗 https://www.framer.com/blog/framer-3/

farza showed a new Clicky demo where the AI can draw directly over your screen, not just talk next to it. using Claude Opus, it marks up a Khan Academy Pythagorean theorem lesson with shapes and pointers, then walks through FL Studio by pointing at the exact rows and controls.

the tutor sees the live workspace and teaches inside it. still a demo, but it’s a strong direction for software learning: guidance that sits on top of the task, not beside it.

🔗 https://x.com/FarzaTV/status/2066983088035656086

Vercel is turning its internal agent infrastructure into an open-source framework called eve. it treats an agent as a folder of files: instructions, tools, skills, subagents, channels, schedules, plus the model config.

why it matters: the boring production pieces are built in, like durable sessions, sandboxed compute, approvals, tracing, evals, and deployment through Vercel. it is public preview and Apache-2.0 on GitHub, so useful to watch, but still beta rather than a fully settled standard.

🔗 https://vercel.com/changelog/introducing-eve-an-open-source-agent-framework

Strawberry is taking a slightly different swing at the AI browser idea. the pitch is not better tabs or cleaner browsing, but agents that live inside the browser your team already uses for research, prospecting, data extraction, and follow-up work.

personal note: that matters because a lot of AI browsers still feel like another layer to manage. after trying Perplexity’s Comet, Atlas, Arc, and Dia, my question is basically: do we need more browser organization, or do we need the browser to quietly do the work?

Strawberry’s bet is the second one, with a free Mac and Windows app live now and a bigger announcement teased for end of summer.

excited to give it a go try! (also... for any heavy browser tasks i personally just use codex rn...)

🔗 https://strawberrybrowser.com
🔗 https://x.com/charles_maddock/status/2067238521023201489

Midjourney’s very unexpected pivot: it announced Midjourney Medical, a full-body ultrasound scanner that lowers you through water and uses a ring of sensors to reconstruct internal body images in about 60 seconds.

the interesting bit is the wrapper, not just the machine. Midjourney wants the first version inside a San Francisco “spa” by end-2027, starting with body-composition maps while diagnostic features still need FDA clearance. very early, but a real swing from an AI image lab.

🔗 https://www.midjourney.com/medical/blogpost

Perplexity is adding Brain to Computer, its agent for getting tasks done. the point is to make Computer less stateless: Brain builds a context graph from past sessions, sources, files, decisions, and corrections, then uses that history to start future tasks with more relevant context.

the important bit is the shift from "memory about you" to memory about the work. Perplexity says early tests on tasks needing past context showed higher correctness and recall, plus lower cost; for now it's a research preview for Max and Enterprise Max subscribers.

i've not tried perplexity in a long time other than for searching similar products; but might give computer a try again and see how this fairs with much larger context

🔗 https://www.perplexity.ai/hub/blog/self-improving-memory-for-agents
🔗 https://x.com/perplexity_ai/status/2067642139014742348

Google is making AI avatars in Google Vids available at no cost for anyone with a Google account in the US, with other regions rolling out later. the new Vids updates turn Slides into narrated videos, add avatars and voiceovers in 24 languages, and support branded avatars for product demos.

the interesting bit is the packaging: Google is moving avatar video from a production tool into the regular Workspace flow, so sales decks, onboarding docs, and internal updates can become short presenter-led videos without setting up a shoot.

🔗 https://x.com/GoogleWorkspace/status/2067249947993366654

Amp added custom agents through plugins.

plugins can now create agents, run them once, keep talking to their threads, and even send messages between agents. useful direction for devtools: agents stop being one-off chat boxes and become programmable workers inside the tool.

🔗 https://ampcode.com/news/custom-agents

Codex Record & Replay has started rolling out on macOS in supported markets.

it lets you record a workflow once, then turn that demo into an inspectable skill Codex can reuse later. Nick’s note says EEA, UK, and Switzerland are not in the first rollout, and that computer use needs to be enabled.

🔗 https://x.com/nickbaumann_/status/2068030706077634604

Cloudflare added temporary accounts for AI agents.

with wrangler deploy --temporary, an agent can deploy to Cloudflare without a user signing up or logging in first. that’s a neat unlock for agentic deploy flows: try the thing first, attach the real account later.

🔗 https://blog.cloudflare.com/temporary-accounts/

Dot Matrix is a tiny site for generating loading animations.

it has 55+ free, open-source loaders, with controls for size, shape, animation, hover effects, bloom, and more. useful little design/dev utility when you need a polished loader without building one from scratch.

🔗 https://dotmatrix.zzzzshawn.cloud/