Meet Your New Coworker: ChatGPT Agent

Hold onto your hotkeys, Tech Insiders.

ChatGPT Agent runs workflows, Slack AI does your thinking for you, and Gemini's auto-summaries blur the lines between friend and foe.

Let's see what's left for us.

Before you dive in:

Hear Grant Harvey and Corey Noles unpack the AI-in-education crisis (or maybe noncrisis?) From UCLA's viral "CheatGPT" scare and MIT's cognitive-debt study to an Austin school where AI tutors condense classes into two hours—come decide for yourself whether bots are wrecking classrooms or reinventing them.

Listen to the conversation on YouTube.  

Here's what you need to know today:

OpenAI's ChatGPT Agent Goes Full 'Jarvis'

OK, who gave the chatbot a laptop and a corporate card?

OpenAI has launched ChatGPT Agent, a unified system that fuses the Operator browser, deep-research mode, and classic ChatGPT into one task-running powerhouse. Pro subscribers get it first, with a 400-message monthly allowance; Plus and Team tiers follow with 40 each month.

The agent can click around the web (in its own visual or text-only browsers), run code in a terminal, call APIs, and churn out editable slides and spreadsheets.

Safety measures include pausing for clarification, asking permission before "consequential" actions, and switching to Watch Mode when sending emails or making purchases. A one-click privacy control wipes browsing data and logs the agent out of every active site session, and new guardrails aim to spot prompt-injection attacks.

OpenAI claims the new model scored record highs on several tough benchmarks—real-world job tasks, advanced math problems, and spreadsheet editing—and matched or beat human test-takers about 50% of the time.

Early users still report latency (it took nearly an hour to order a large cupcake shipment) and note that slideshow formatting feels "beta." Access for Enterprise/Education users is coming later this summer, while the European Economic Area has no set deadline yet.

Why it matters: If it works as billed, watching a bot book flights, update sheets, and draft decks could turn AI from chat sidekick into full-fledged digital colleague, saving clicks while adding fresh security questions (and a handy scapegoat when the cupcakes show up late).

Share this newsletter!

Pulse Check

Will you let ChatGPT Agent tackle your day-to-day tasks?

Results from Yesterday's Pulse Check

Would you let Google's AI handle your next service call?

Insider Intel

Slack Supercharges Workflows With New Native AI

The office watercooler just learned NLP.

Slack is weaving AI straight into its chat fabric. Its new enterprise search pulls answers from connected apps—Google Drive, Microsoft Teams, Salesforce, Confluence, and more—so you can surface buried files with one prompt. Fall behind in a channel, and auto-generated recaps catch you up. Need to bridge a language gap? Real-time translations keep conversations flowing in any tongue. And huddles get instant meeting notes.

Coming "later this year," writing assistance in Canvas will turn brainstorms into polished briefs (and even rewrite tone). AI-powered action items will flag your to-dos directly from conversation context. AI profile summaries will help newcomers figure out who's who. And message explanations will let you hover over any cryptic acronym for an instant decoder-ring translation.

The rollout is tiered:

Slack says customer data never trains external models, and AI results respect each user's existing permissions.

So yes, Slack can now answer "Where's that deck?" before you finish typing—though it still can't stop colleagues from pinging you with "Quick question?" at 16:59 every day.

Flash Bytes

How to Build GenAI Apps with CockroachDB & AWS Bedrock

Discover how to build next-generation GenAI applications with scalable, resilient architecture.

Join How to Build GenAI Apps with CockroachDB & AWS Bedrock on August 6, 2025, at 1:00 PM ET, hosted by Cockroach Labs. Learn how vector search, RAG, and real-time AI models can reshape your deployments.

Security Alerts

Gemini Email Summaries Open a Phishing Backdoor

Researchers found Google's Gemini for Workspace can be tricked by invisible HTML instructions—white-on-white text hidden inside an email. When a user clicks "Summarize this email," Gemini obediently adds the attacker's fake security alert, urging victims to call a number or click a link. No attachments, no obvious red flags, just prompt-injection sleight of hand.

Google says new layered defenses—content classifiers, URL redaction, and human-in-the-loop confirmations—are rolling out, but the proof of concept still works today. Until patches land, security teams should scrub zero-font content and remind staff: AI summaries are helpful, not gospel.

Pro tip: If Gemini warns your "password is compromised," yet you never saw the message, pause before dialing that 1-800 number.

Salt Typhoon's Nine-Month Guard Infiltration Exposed

A DHS memo reveals China-linked APT Salt Typhoon camped inside one state's Army National Guard network for nine months, swiping admin credentials, network diagrams, and service-member PII.

The 2024 breach, which spanned from March to December, also monitored traffic with Guard units in every state and at least four US territories, giving Beijing a potential blueprint to hobble local defenses in a crisis.

Investigators say the group exfiltrated 1,462 configuration files tied to roughly 70 government and critical-infrastructure entities, leveraging Cisco and other device CVEs to move laterally. Officials insist missions weren't disrupted, but analysts warn that mapping Guard–fusion-center links hands attackers a ready-made playbook for disrupting critical infrastructure.

Look on the bright side—at least someone finally documented those network diagrams.

Industry Shakeups

Amazon Trims AWS Staff After 'Strategic Review'

Amazon has laid off hundreds across AWS, including customer-facing specialists, following a strategic review. The announcement landed just one day after an official AWS blog post touted the "unprecedented opportunities" AI is creating for early-career talent.

The company maintains that AI wasn't the primary trigger, despite CEO Andy Jassy's June memo warning that AI efficiencies would eventually reduce some roles.

Affected US employees will receive at least 60 days of pay and benefits plus job-search support, and Amazon says hiring continues in priority areas.

The cuts arrive even as AWS reported solid Q1 results: $29.3 billion in revenue (up 17% year over year) and $11.5 billion in operating income (up 23%). Still, that 17% top-line increase marked the division's slowest growth in more than a year, pressuring leadership to streamline while plowing cash into strategic bets like generative-AI services.

Software may be eating the world, but apparently it skipped leg day and went straight for middle management.

Advertise in Daily Tech Insider!

Daily Tech Insider is a TechnologyAdvice business

© 2025 TechnologyAdvice, LLC. All rights reserved.

TechnologyAdvice, 3343 Perimeter Hill Dr., Suite 100, Nashville, TN 37211, USA.