In partnership with
Talk to your AI tools the way you'd talk to a colleague.
You don't send a colleague a three-word brief. You explain the context, the constraints, what you've already tried. But typing all that into ChatGPT takes forever — so you don't.
Wispr Flow lets you speak your prompts instead. Talk through your thinking naturally and get clean, paste-ready text. No filler words. No cleanup. Just detailed prompts that actually get you useful answers on the first try.
Millions of users worldwide. Works system-wide on Mac, Windows, and iPhone.
Google I/O 2026: The Search Giant Just Rebuilt Everything Around Agents
Google I/O 2026 wasn't a developer conference. It was a product reset. Every major announcement pointed in the same direction: agents are the new interface, and Google is rebuilding its entire ecosystem around them.
The headliner was Gemini 3.5 Flash, Google's newest model that combines what the company calls "frontier intelligence with action." It outperforms Gemini 3.1 Pro on coding and agentic benchmarks while running 4x faster on output tokens. On Terminal-Bench 2.1, it scores 76.2%. On MCP Atlas, 83.6%. This is a model built to do things, not just answer questions.
But the model wasn't the real story. The real story was everything Google built on top of it.
Search just changed forever
Google called it the biggest change to Search in 25 years. AI Mode, already used by over 1 billion people monthly, is now powered by Gemini 3.5 Flash globally. But the real shift is structural: instead of returning a list of links, Search can now build custom generative UIs on the fly, complete with interactive tools, visualizations, and simulations tailored to each query.
Google also announced that users will be able to create personal "information agents" directly inside Search. These agents work in the background 24/7, monitoring the web for changes and alerting you when something relevant happens. Think of it as Google Alerts rebuilt for the agent era.
Gemini Spark: your always-on AI employee
The most significant product announcement was Gemini Spark, a personal AI agent that runs on dedicated virtual machines in Google Cloud. You can email Spark directly through a dedicated Gmail address, and it will execute tasks without you needing to keep your laptop open. It browses the web through Chrome, works across Workspace apps, and connects to third-party services including Canva, OpenTable, and Instacart.
This isn't a chatbot. It's a persistent digital worker that runs 24/7 in the cloud. The model behind it: Gemini 3.5 running on Google's Antigravity agent platform.
The developer platform shift
Google also upgraded Antigravity with Managed Agents. A single API call now provisions a remote Linux environment where the agent can reason, plan, call tools, execute code, manage files, and browse the web. For developers, this is the equivalent of spinning up a full AI employee in one line of code.
And on the creative side: Gemini Omni, a new model series, generates and edits video from text, image, audio, and video inputs. Adobe, Canva, and CapCut are all integrating directly into the Gemini app.
Google isn't competing on model benchmarks anymore. It's competing on infrastructure: agents that run in the cloud, work across apps, and never stop. If your team is evaluating AI platforms, this week changed the calculus. Gemini Spark alone means any Google Workspace customer can now deploy a persistent AI agent without building anything. The question isn't whether AI agents will become standard productivity tools. It's whether your team will adopt them before your competitors do.
Google Spark, Microsoft Copilot Cowork, OpenAI's super app. Every major platform is shipping persistent AI agents this year. Before your team deploys any of them, answer these five questions.
Q1: Which workflows are actually repetitive?
Agents excel at predictable, repeatable tasks. They struggle with ambiguous, judgment-heavy decisions. Start by listing tasks where the steps rarely change: inbox triage, meeting scheduling, report formatting, data entry, status updates, research monitoring, calendar management.
The test is simple. If a task requires explaining new context every time, it's not agent-ready yet. If a task has clear rules and predictable inputs, it's your first candidate.
Q2: Can the agent access what it needs?
A persistent agent is useless if your critical information lives in disconnected systems. Audit where your team's knowledge actually lives: email threads, Slack channels, shared drives, CRM notes, project management tools, internal wikis.
Gemini Spark integrates with Workspace. Copilot Cowork integrates with 365. The agent is only as good as the systems it can reach. If the data it needs is trapped in silos, fix the access problem before you deploy.
Q3: Who reviews the output?
No agent should run fully autonomous on day one. Every deployment needs a human review layer. Define before you launch: who checks the agent's work, how often they check, what the escalation path looks like when something goes wrong, and what gets auto-approved versus what needs a human sign-off.
The teams that skip this step are the ones that end up with agents sending wrong emails, booking conflicting meetings, and filing incorrect reports. Design the review loop before the agent starts working. Not after the first mistake.
Q4: What's the failure mode?
Not all tasks carry the same risk. The answer to this question determines your entire deployment strategy.
Drafting meeting summaries, organizing files, monitoring news feeds, compiling research digests, formatting status updates. Let agents run these with periodic spot-checks.
Client email drafts, calendar scheduling with external contacts, data analysis for internal reports. Agent does the work, human approves before it goes live.
Financial calculations, legal document preparation, external communications on behalf of the company, anything involving customer-facing commitments. These need multi-step human approval and audit trails.
Map every candidate workflow to a risk tier before you deploy. The escalation path for a low-risk task is a weekly review. The escalation path for a high-risk task is real-time human approval on every output.
Q5: How do you measure success?
"We deployed an AI agent" is not a success metric. Without clear measurements, you'll never know if the agent is helping or just generating activity that feels productive.
Define measurable outcomes before launch. Here's a starting framework:
Time saved: Hours per week recovered from the automated workflow.
Error rate: Percentage of outputs requiring human correction.
Throughput: Tasks completed per day versus the pre-agent baseline.
Response time: How fast the agent completes work versus the old process.
Adoption: How many team members are actually using the agent weekly.
Benchmark these numbers before the agent starts. Measure again at 30 and 90 days. If the numbers aren't moving, the deployment isn't working, regardless of how impressive the demo looked.
Pick one task your team does every week that follows the same steps. Ask three questions: can the agent access the tools it needs, who will review the output, and how will you know it's working? If you can answer all three clearly, that's your first agent deployment candidate. Time required: 10 minutes. The best deployments start small and expand fast. The worst ones start big and collapse slowly.
Anthropic Hits $10.9B Projected Q2 Revenue, Posts First Operating Profit
Anthropic shared financial projections showing $10.9 billion in revenue for Q2 2026, up 130% from $4.8 billion in Q1. The company expects approximately $559 million in operating income, marking its first ever quarterly profit. For context, Anthropic is also paying SpaceX $1.25 billion per month for compute access through 2029. The fact that they're still profitable after that expense says everything about demand for Claude.
Andrej Karpathy Leaves OpenAI's Orbit, Joins Anthropic
Andrej Karpathy, one of OpenAI's founding members and former head of Tesla's Autopilot program, announced he's joining Anthropic to lead a new team focused on using Claude to accelerate pre-training research. This is arguably the highest-profile talent move in AI since the founding of the major labs. Karpathy brings deep expertise in both language models and applied AI systems. Anthropic just acquired one of the most respected AI researchers alive.
Jury Dismisses All of Musk's Claims Against OpenAI in Under Two Hours
A California federal jury unanimously rejected every claim Elon Musk filed against OpenAI and Sam Altman, deliberating for less than two hours after eleven days of testimony. The jury found that Musk waited too long to file, barring all claims under the statute of limitations. If Musk had won, OpenAI and Microsoft could have been forced to return up to $150 billion to OpenAI's nonprofit foundation. Musk called it a "calendar technicality" and vowed to appeal.
NextEra Acquires Dominion Energy for $67B: The Largest Utility Merger in US History
NextEra Energy announced a $67 billion all-stock deal to acquire Dominion Energy, which powers Northern Virginia, the world's largest concentration of data centers. The deal is explicitly driven by AI power demand. Data centers are projected to consume 15-25% of US electricity by 2030. This is the clearest signal yet that AI infrastructure isn't just a tech story. It's reshaping the energy sector.
Cursor Composer 2.5 Matches Claude Opus 4.7 at One-Tenth the Cost
Cursor launched Composer 2.5, built on Moonshot AI's open-source Kimi K2.5 (1 trillion total parameters, 32 billion active). It scores 79.8% on SWE-Bench Multilingual, essentially tied with Opus 4.7 at 80.5%. Standard pricing: $0.50/$2.50 per million tokens, roughly 10x cheaper than Opus. The era of frontier-quality coding at commodity prices just arrived. Cursor is already training a larger successor model on SpaceX's Colossus 2 supercomputer.
Gemini Spark
What it is: A persistent AI agent that runs on dedicated virtual machines in Google Cloud. Unlike every other AI assistant, Spark doesn't need your laptop open to work. You can email it tasks through a dedicated Gmail address, and it executes them autonomously: browsing the web through Chrome, working across Workspace apps, and connecting to third-party services.
Why it matters now: This is the first consumer-facing AI agent that truly runs 24/7 without human supervision. It's built on Gemini 3.5 and powered by Google's Antigravity agent platform. The integrations at launch include Google Workspace (Gmail, Docs, Slides), Canva, OpenTable, and Instacart. Google says it will expand to all Gemini-connected apps.
Who gets it: Spark is currently in testing at Google internally. It's expected to roll out to Google AI Ultra subscribers next week. If you're already on the Ultra plan, this is a significant upgrade at no additional cost. For teams evaluating whether to standardize on Google's ecosystem, Spark may be the deciding factor.
"I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D."
Andrej Karpathy, OpenAI co-founder, announcing his move to Anthropic · May 19, 2026
Know someone navigating AI decisions? Forward this issue.
SHARE & EARN REWARDS →aiinsider247.com
You're receiving this because you subscribed to AI News Insider.
Unsubscribe · View in browser