A general-purpose work agent reached millions of paid ChatGPT users, ten of the biggest enterprise vendors lined up behind a new standard for connecting agents to business software, Meta opened its agent platform worldwide, and regulators in China and the EU both made moves that change how agents get deployed and distributed. Here is what happened and why it matters.
OpenAI's ChatGPT Work agent rolls out, alongside the GPT-5.6 family
OpenAI launched ChatGPT Work on July 9 and it is now reaching Plus and Business plans. You give it a goal, it pulls context from connected apps and files, works across multiple steps for hours, and hands back finished documents, spreadsheets, and reports. It can also run on a schedule or on triggers.
It shipped alongside the GPT-5.6 family: flagship Sol at $5/$30 per million tokens, mid-tier Terra, and high-volume Luna at $1/$6. With xAI's Grok 4.5 at $2/$6 and Moonshot's open-weight Kimi K3 arriving the same week, the cost of running production agent workloads keeps dropping. For businesses, the real question is no longer whether an off-the-shelf agent can do useful work. It is which tasks deserve a general-purpose assistant and which need a purpose-built deployment with controls around it.
Big vendors back a rival standard for wiring agents into business software
Google, Microsoft, Salesforce, Snowflake, ServiceNow, Cisco, Databricks, GitHub, NVIDIA, and Hugging Face are jointly building a federated discovery standard that lets an agent find the right enterprise resources without custom connectors. It is positioned as an alternative to Anthropic's Model Context Protocol, which has been the default answer for a year.
A standards fight among vendors this large means one thing for anyone deploying agents now: build integrations you can swap. Betting your whole stack on a single protocol before this settles is a risk you do not need to take.
Meta opens its Business Agent Platform worldwide
Meta made its Business Agent Platform available globally, letting companies build and deploy agents across WhatsApp, Messenger, and Instagram. It launched alongside Meta Compute, which sells Meta's surplus AI infrastructure in direct competition with AWS, Google Cloud, and Azure.
If your customers live on WhatsApp or Instagram, this puts a serious agent channel where they already are. It also adds a fourth heavyweight to the enterprise-agent platform race, which means more choice and more pressure to pick deployments you can actually trust and move if needed.
China's dedicated AI-agent rules take effect, a global first
China's Implementation Opinions on intelligent agents became enforceable on July 15, creating the world's first regulatory category built specifically for AI agents. The rules set a three-tier framework for how much decision authority an agent may hold and require filings for deployments in high-risk sectors.
If you operate in or sell into China, agent autonomy is now a compliance question with formal answers. Even if you do not, other regulators tend to study templates like this one, so the way these rules classify agent authority is worth understanding early.
The EU orders Google to open Android to rival AI assistants
The European Commission adopted two binding Digital Markets Act decisions on July 16 requiring Google to open 11 Android features, including voice activation and cross-app actions, to third-party AI assistants, and to share anonymized search data with competitors on a phased schedule into 2027.
For companies building or deploying agents in Europe, a major distribution barrier just got lower. Assistants and agents that are not Google's will be able to plug into the phone in ways only Google could before.
Anthropic's deputy CISO publishes a four-question test for agent risk
Jason Clinton, Anthropic's deputy CISO, published a practical framework for approving agent deployments. It comes down to four questions: how trusted is the content the agent ingests, what actions is it allowed to take, what is the blast radius if it goes wrong, and how observable is its behavior.
This is the most useful piece of agent-security writing we have seen in a while, because it gives security teams a way to say yes with conditions instead of blocking pilots outright. If you are trying to get an agent project past IT or compliance, these four questions are a good shared language.
What this means for you
The platforms are converging on the same idea from different directions: agents that do real work inside your business, governed by real rules. Off-the-shelf agents like ChatGPT Work will absorb a lot of general tasks. The work that touches your customers, your money, and your data still needs deliberate deployment choices: which platform, which protocol, what the agent is allowed to do, and who stands behind it when something breaks. The companies that answer those questions deliberately will get the value; the ones that improvise will get the incidents.
Want to know where agents fit in your business?
Book a free discovery call. We'll map your workflows to the agents worth deploying, and match you with the vetted engineer to build them.
Book a Discovery Call