Google wants agents to take actions

Google turns AI agents into tool users

Google just gave its AI agents a long list of logins.

The company expanded its open source Agent Development Kit with direct integrations to platforms like GitHub, GitLab, MongoDB, Stripe, PayPal, and Hugging Face. With a few lines of code, developers can connect an agent to tools for code execution, payments, databases, observability, and even email through providers like Mailgun.

This is Google pushing agents from chat into operations. An agent that can open a pull request, query a vector database, or send an invoice through Stripe is much closer to replacing real workflow software.

The partner list is broad, from Atlassian and Notion to Pinecone and Weave, and it sits alongside built in Google Cloud services like BigQuery and Pub/Sub. Picture a developer at 10 p.m. with a laptop open, wiring an agent into GitHub and Asana in under an hour.

If this sticks, the real competition shifts from model quality to ecosystem depth and who controls the pipes agents run through.

Read AI turns your inbox into an agent

Read AI wants to sit in your inbox and act like you.

The company launched Ada, an email based assistant that can schedule meetings, draft replies, answer questions from your company’s knowledge base, and pull context from past meetings. Users activate it by emailing [email protected], and it can negotiate times over email without exposing details from your calendar.

This is a smart wedge. Email is still where scheduling chaos and status updates live, and Read AI already has access to meeting data that gives it useful context.

The company says it has 5 million monthly active users and sees 50,000 new sign ups a day, which gives it distribution most AI startups would kill for. Picture the morning inbox flood at 8:30 a.m., and an assistant quietly replying with time slots and draft answers before you take a sip of coffee.

If Ada works reliably, it shifts Read AI from note taker to workflow layer. The risk is trust, because one wrong reply can cost more than a missed meeting.

OpenAI moves into the Pentagon’s AI stack

OpenAI is now inside the Pentagon’s classified network.

After Anthropic’s relationship with the Defense Department collapsed over concerns tied to surveillance and autonomous weapons, OpenAI reached its own agreement late Friday, according to Sam Altman’s post on X. Altman said the deal reflects OpenAI’s principles, including prohibitions on domestic mass surveillance and a requirement for human responsibility in the use of force, and that safeguards were built into the deployment.

This is a clear power shift. The Pentagon still wants advanced AI models in secure systems, and OpenAI is signaling it is willing to work with the military as long as it can publicly define the guardrails.

The timing matters. Late Friday announcements are rarely casual, and this one suggests the competition for defense AI contracts is heating up behind closed doors. If OpenAI becomes embedded in classified workflows, it will shape how military teams use generative models day to day.

The real question is whether those principles hold once the systems are fully in use.

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.