AI & Automation
AI that keeps working: what the latest agent launches mean for business
9 October 2026 · 6 min read
A single week of AI announcements — from OpenAI, Anthropic, Google, Perplexity and Microsoft — had one theme running through all of it: AI is moving beyond answering prompts and towards using software, monitoring work and taking responsibility for ongoing tasks. Here is what was announced, and what it means if you run a business rather than a research lab.
OpenAI's Dots: AI workers that run in the cloud
OpenAI introduced Dots, ChatGPT AI workers that can run in the cloud, use connected apps and handle ongoing responsibilities — asking for approval when a decision needs a human.
That last detail is the important one. The pattern isn't an unsupervised bot; it's delegation with checkpoints, which is the same model good operations teams already use with people.
Shared workspaces and new models
ChatGPT also gained Spaces, where teams, ChatGPT and Dots collaborate on shared documents and presentations. Alongside it came two models with different priorities: GPT-6.1 Sol, positioned as cheaper for long-running work, and Astra Ultra, which prioritises speed.
The split reflects a real choice businesses will face: an agent that works quietly in the background and an assistant that answers instantly have different cost and latency needs, and rarely belong on the same model.
AI building AI: six months of work in under 30 days
OpenAI reported that merging ChatGPT's web and desktop codebases — a project expected to take six months — was completed in under 30 days with AI assistance.
Treat vendor-reported numbers with healthy scepticism, but the direction is consistent with what engineering teams are seeing: large, tedious migrations are exactly where AI assistance changes the schedule.
Codex moves to cloud-based agent work
Codex expanded into cloud-based agent work. Coding tasks can run remotely, scan repositories for security issues and use computers to test software, and they can be managed through new voice, review and multi-agent features.
ChatGPT itself became more connected and proactive too: plugins, app-triggered automations, recurring tasks, meeting notes, sign-in integrations and a marketplace let it act across tools and workflows rather than waiting in a chat window.
Anthropic, Perplexity and Google
Anthropic updated its everyday model with Claude Sonnet 5.5, presented as faster and cheaper than its predecessor, with improvements in coding, knowledge work, computer use and visual understanding.
Perplexity added proactive automations: its Computer can monitor connected apps, respond to changes, run scheduled tasks and remember prior work without using credits while idle.
Google's Gemini 4 Argon is built for longer tasks, able to produce up to one million output tokens. Google says it helped update a large codebase and make a video system 2.7 times faster.
Smaller updates with practical reach
Perplexity can now create interactive visual answers. Microsoft introduced faster speech models for more natural conversations. Google launched a personalised health coach.
None of these is the headline, but together they show the same pattern: AI features are becoming part of everyday tools, not a separate destination you visit.
What this means for your business
The common thread is persistence. Instead of answering one question and stopping, these systems monitor, wait for triggers, run on a schedule, remember context and come back for approval. That is closer to a junior team member than to a search box.
It makes the groundwork more important, not less. An agent is only as useful as the systems it can reach, the processes it can follow and the guardrails around what it may do. Businesses with clean workflows, integrated systems and clear approval points will get value from these tools quickly; those with tangled spreadsheets and undocumented workarounds will not.
The sensible sequence hasn't changed: map the workflow, automate the mechanical steps, apply AI where judgement is genuinely needed, and keep a human approval step on anything consequential.
The shift is from AI that answers to AI that keeps working — and it rewards the businesses whose systems are ready for it.
Questions
- What is an AI worker like OpenAI's Dots?
- As announced, Dots are ChatGPT AI workers that run in the cloud, use connected apps and handle ongoing responsibilities, asking for human approval when needed — closer to a delegated team member than a one-off chatbot.
- What does 'AI that keeps working' actually mean?
- It means AI that doesn't stop after one answer: it monitors connected apps, reacts to changes, runs scheduled tasks and remembers earlier work, rather than waiting for a new prompt each time.
- Should my business adopt AI agents now?
- Start with a well-defined, low-risk workflow with a clear approval step. Agents amplify the quality of the systems and processes underneath them, so integration and process clarity matter more than which model you pick.
- Are these AI announcements and benchmarks reliable?
- Figures such as development timelines and speed-ups are reported by the vendors themselves. Use them as direction, and evaluate any tool against your own workflows and data before relying on it.