The Nasdaq 100 completed a 10% drawdown in just 38 trading days, led by memory: SK Hynix 8.98%, Micron 8.85%, SanDisk 14%. Meanwhile open models keep shipping faster than the norms around them.
Zhiyuan/AgiBot's WITA-Omni Preview topped the DailyOmni multimodal benchmark on July 28 with an 85.21 composite, first in 6 of 8 sub-metrics. It beat Gemini 3.1 Pro Preview, Qwen3.5-Omni-Plus and Doubao Seed 2.0 Lite.
The Nasdaq 100 fell 10% from its high on Tuesday, finishing the correction in 38 trading days. SanDisk sits more than 50% below its high set a month ago, and OPEC+ signalled a production freeze after September.
Amazon shipped Kiro IDE 1.0, Claude Cowork arrived on web and mobile, and Anthropic released Reflect for usage audits. The agent tooling layer is consolidating into products you can actually govern.
Koda publishes a fact-checked AI intelligence briefing every morning: site, email, podcast, and video. Founding sponsors lock today's rates for 6 months as the audience compounds.
View the media kit →Instagram began banning accounts posting harassment videos shot on Meta glasses, targeting pickup-artist and prank content filmed on unsuspecting strangers in public. The Verge argues Meta is mishandling the rollout entirely: the hardware shipped without norms or enforcement in place, and the company is now...
Continue reading arrow_outwardInstagram started banning accounts posting harassment videos filmed on Meta glasses, moderation policy arriving after the hardware shipped rather than before.
A magnitude 7.1 earthquake struck Japan's Kumamoto region on 28 July, killing multiple people, collapsing houses and drawing more than 100 aftershocks that hampered search efforts.
Mood is Fear: the Nasdaq 100 is 10% off its high with SK Hynix down 8.98%, Micron 8.85% and SanDisk 14%, even as Apple briefly touched a $5 trillion valuation.
NVIDIA spent 27 July recruiting Adobe, Dell, CrowdStrike and Hugging Face into a security alliance for open models, and the next day a Chinese open-lab entrant, WITA-Omni Preview, took the DailyOmni crown from Gemini 3.1 Pro: the safety scaffolding and the competitive pressure are both building around open weights.
Meta's glasses are now a moderation problem: Instagram is banning pickup-artist and prank accounts that filmed unsuspecting strangers in public. The Verge's critique is structural rather than incidental, arguing the device went out without enforcement in place and the policy is being retrofitted onto a deployed product. That sequence, ship first and govern later, is the pattern to watch as more always-on cameras reach consumers.
The Nasdaq 100's 10% drop in 38 sessions was driven by the AI hardware complex, not the index broadly: SK Hynix fell 8.98%, Micron 8.85%, and SanDisk 14%, leaving SanDisk more than 50% below a high set only a month ago. Apple briefly crossing $5 trillion in the same session shows how concentrated the remaining strength is. When the supply chain sells off while the largest cap sets records, the trade is narrowing rather than broadening.
NVIDIA's Open Secure AI Alliance, launched 27 July with Adobe, Dell, CrowdStrike and Hugging Face, is an argument that open weights can be secured with shared tooling. A day later, Zhiyuan/AgiBot's WITA-Omni Preview scored 85.21 on DailyOmni, first in six of eight sub-metrics and ahead of Gemini 3.1 Pro Preview, Qwen3.5-Omni-Plus and Doubao Seed 2.0 Lite. Another Chinese lab holding a public benchmark crown this month raises the stakes on who sets open-model security standards.
Kiro hit 1.0 with a rebuilt interface organized around telling an agent what to build rather than typing every line yourself. Treat it as a spec-first editor: write the requirement, let the agent scaffold, then review diffs before accepting. Worth a side-by-side trial against your current editor on one small feature branch this week.
Try it arrow_outwardAnthropic is extending Claude Cowork beyond the desktop so agentic work sessions continue across devices and keep running in the background without your machine staying online. Kick off a long refactor or research sweep from your laptop, then check progress from your phone. Best fit for tasks measured in hours, not seconds.
Try it arrow_outwardReflect is a beta tool that charts your Claude usage over the past 1, 3, 6, or 12 months so you can see where the assistant actually earns its keep. Run a 6-month view and look for two things: tasks you delegate constantly (candidates for automation or a saved prompt) and tasks you delegate that you should still be doing yourself.
Try it arrow_outwardMeta AI now pulls reminders, tasks, and calendar into one place and can connect to email and calendar apps to build plans and track activity. It is the lightest-weight option if you already live in Meta's apps and do not want another standalone planner. Start by connecting one calendar only, then judge whether the plans it drafts are worth widening access.
Try it arrow_outwardDevin Review is listed on There's An AI For That as a review-layer companion to agentic coding, aimed at the growing pile of machine-written pull requests nobody has time to read closely. Point it at one recent agent-generated PR and compare its findings against your own read before you wire it into CI. Check current pricing on the listing; terms are not published in the directory entry.
Try it arrow_outwardPixo is an image generation entry surfacing on There's An AI For That this week, with alternatives and pricing viewable directly on its listing page. Use it for throwaway visual drafts, slide art, and mood boards where you need volume rather than a final asset. Compare output on one prompt against whatever you use now before switching anything in your workflow.
Try it arrow_outwardPin one project to Kiro IDE 1.0. Take a repo you already know and run a spec-driven task end to end, then log where the agent needed a human correction. That log is your real benchmark, not a leaderboard.
Run a usage audit with Anthropic Reflect. Point it at last month's agent traffic and separate genuine automation from retry loops and abandoned sessions. Cut the top two waste sources before you add capacity.
Stress test an open-weight model against your task. Pull WITA-Omni Preview or Qwen3.5-Omni-Plus and score them on your own multimodal evals rather than DailyOmni's 85.21 composite. Then decide if a closed API still earns its price.
Meta's glasses became a moderation problem before the norms existed, and Instagram is now banning accounts retroactively. NVIDIA's Open Secure AI Alliance, launched July 27 with Adobe, Dell, CrowdStrike and Hugging Face, is the opposite bet: secure open weights with shared tooling before the incident, not after. The market is pricing the same tension, with a 10% Nasdaq 100 drawdown in 38 sessions and SanDisk more than 50% off its high. Builders who instrument and audit their own stacks this quarter will be the ones still shipping when the capital tightens.
AI analysis, world news, markets, and tools. One briefing, delivered free.