What is ChatGPT Work?

ChatGPT Work is OpenAI's autonomous workplace agent, launched on 9 July 2026. Instead of returning a single response, it carries out a full assignment: it pulls context from your connected apps and files, coordinates multiple actions over an extended period, and delivers a finished output — a website, a report, a spreadsheet, a presentation.

It is available across ChatGPT on web, mobile and desktop.

The distinction matters. A chatbot shortens the time it takes to write something. An agent shortens the time it takes to finish something.

What is GPT-5.6? Sol, Terra and Luna explained

GPT-5.6 is OpenAI's model generation released alongside ChatGPT Work on 9 July 2026. It ships as three tiers. The number is the generation; the name is the capability tier.

GPT-5.6 is available through ChatGPT, Codex and the OpenAI API, and is already live in GitHub Copilot.

Is this the same as Claude Cowork?

Broadly, yes — and Anthropic shipped first.

Claude Cowork launched as a research preview on 12 January 2026 for macOS, expanded to Windows on 10 February, reached general availability across all paid plans on 9 April, and rolled out to web and mobile on 7 July 2026 — two days before OpenAI's announcement.

Both products solve the same problem: an AI agent that works inside your files and applications to complete real assignments. Coverage of the OpenAI launch has consistently framed it as arriving months after Cowork.

The competitive point for business owners is simple. This is no longer one company's experiment. It is now the direction of the category, and it is being built by the two largest AI labs simultaneously. That makes it a planning assumption, not a novelty.

Do these agents work without human supervision?

No — and this is the single most over-claimed part of the coverage.

OpenAI's own documentation states that ChatGPT requests permission before taking actions of consequence, and that users can interrupt the agent, take over the browser, or stop a task at any point.

The accurate description is extended autonomy with human checkpoints, not unsupervised operation. Any vendor or consultant promising you a fully unsupervised AI workforce is describing something the product does not do.

What this actually changes for South African businesses

1. It lowers the cost of execution, not the cost of strategy

Most South African SMEs are not short of ideas. They are short of hours. The competitor analysis nobody had time to compile. The customer database nobody cleaned. The monthly report that takes a full day and therefore happens quarterly.

Agents attack that backlog directly. The strategic question — what is worth doing — remains a human one, and becomes more valuable, not less.

2. It rewards businesses with clean inputs

These agents operate on your connected files and systems. If your folder structure is chaotic, your CRM half-populated and your processes undocumented, an agent will produce disorganised work faster than a person would.

The preparation work is unglamorous and it is the whole game: organise your data, document your processes, tidy your systems. Businesses that do this before adopting agents will pull ahead of those that adopt first and organise later.

3. Load shedding and connectivity still apply

Both products are cloud-dependent for their heaviest work. South African operational realities — power interruptions, variable connectivity, data costs — remain part of the planning. Agent workloads that run for hours need infrastructure that stays up for hours.

4. POPIA does not pause for AI

Pointing an autonomous agent at files containing customer personal information is a processing activity under POPIA. Before connecting an agent to client data, confirm your lawful basis, your data-sharing terms with the provider, and your internal access controls. Speed is not a defence.

5. The differentiator moves from access to judgement

When every competitor has the same agent for roughly the same monthly fee, the tool stops being an advantage. What remains scarce is knowing which problems are worth solving, which outputs are actually good, and where the agent is confidently wrong.

What South African businesses should do next

• Audit before you automate. List the recurring tasks that consume the most hours and produce the least judgement. That is your adoption shortlist.
• Clean your data. Folder structure, naming conventions, CRM hygiene. This is the prerequisite, not the follow-up.
• Start with reversible work. Internal reports, research summaries, first drafts. Not client-facing communication and not financial transactions.
• Keep a human reviewer. Assign a named person to check agent output before it leaves the business. Every time.
• Measure it. If you cannot show hours saved or output increased after 90 days, you have bought a subscription, not a capability.

Frequently asked questions

Classic Fusion Media  |  CFM-ART-2026-001  |  ChatGPT Work & GPT-5.6 Briefing