Work Gets Delegated. AI Gets Embedded. Platforms Win.

What happened
OpenAI launched ChatGPT Work, an agent inside ChatGPT that can pull information from apps and workflows, create deliverables like sheets, slides, docs, and web apps, and keep long-running projects moving with scheduled tasks; OpenAI says the rollout starts today on web and mobile for Pro, Enterprise, and Edu, while desktop access to Chat, Work, and Codex is available across all plans.

Why it matters
This is a meaningful step away from “ask-answer” chat and toward software that can stay on a task, move across tools, and turn a goal into a finished work product inside real business workflows.

What’s next
The near-term watch item is distribution: OpenAI says Plus and Business plans get web and mobile rollout over the next few days, which will test whether this becomes a premium power tool or a default way mainstream users manage work.

What happened
IBM updated IBM Bob, its agentic software-development platform, with multi-agent capabilities, built-in AI cost and usage analytics, and prebuilt modernization workflows for IBM Z, IBM i, and Java environments.

Why it matters
Enterprise AI bottlenecks are moving beyond code generation and into review, validation, modernization, and governance; IBM is betting the next winning product is not a better coding copilot, but an orchestrated delivery layer with cost controls and auditable workflows.

What’s next
IBM is making the latest Bob version available now and pairing it with premium modernization packages, so the next signal to watch is whether regulated enterprises adopt these workflows as a default path for legacy-system rewrites.

What happened
OpenAI announced GPT-5.6 will become the preferred model in Microsoft 365 Copilot, bringing the new model series into Word, Excel, PowerPoint, Chat, and Cowork for Microsoft 365 customers.

Why it matters
The big enterprise story is not just model release cadence anymore; it is default distribution inside existing software suites. When a frontier model becomes the standard layer in office tools, AI adoption moves from experimentation toward embedded workflow.

What’s next
Microsoft says customers will get GPT-5.6 across core productivity apps, while OpenAI notes Microsoft will access the model both natively and through the API, pointing to deeper product coupling between model vendors and software incumbents.

What happened
TechCrunch reported that Meta is on track to begin making the latest versions of its AI-specific chips in September, with Broadcom on design and TSMC on manufacturing, as the company tries to lower GPU costs during a component shortage.

Why it matters
This is what AI vertical integration looks like in practice: the biggest labs are no longer just buying compute, they are designing it. That matters because model economics increasingly depend on who controls inference cost, training capacity, and supply-chain access.

What’s next
Meta intends to use the chips for ranking and recommendation models, broader AI workloads, and inference for its apps, while still spending heavily on outside providers, so the next phase is whether custom silicon meaningfully bends its compute-cost curve.

What happened
WIRED reported that 1X revealed new five-finger hands for its Neo home robot, using tendon-like actuators, 25 degrees of freedom, slip detection, and fast finger motion aimed at handling irregular objects and fine manipulation tasks.

Why it matters
Physical AI does not scale on demos alone; it scales when robots can manipulate the messy objects that fill homes and workplaces. Better hands are not cosmetic—they are a prerequisite for moving from “watch this” robotics to actually useful embodied systems.

What’s next
1X says full automation remains the goal, but WIRED notes Neo is still partly teleoperated today and early-access deliveries are prioritized for 2026, so the real test is whether this dexterity holds up in autonomous, everyday household use.

AI is moving beyond being a tool you open and becoming the system that moves work forward. As agents become embedded in everyday software and physical systems, the competitive advantage shifts from using AI to redesigning how work gets done.

This week, delegate a real project instead of asking a single question. Use ChatGPT Work to research a topic, create a presentation or spreadsheet, and schedule a follow-up task. Then ask yourself:

Did I use AI as a chatbot—or as a teammate?