Key takeaways: If your engineers use Codex, they are in good hands. OpenAI builds it for coding: its cloud product is described as running coding tasks in parallel cloud environments, and its Slack app is for kicking off coding work. The question this page answers is the one founders and ops leads actually ask: does that cover the rest of the team? For everyday non-coding work OpenAI points people to a sibling product, ChatGPT Work, which lives inside ChatGPT. Bob is a different shape: one AI employee the whole team shares in Slack, who does the work that wastes your day, asks before anything sensitive goes out, and shows what every job cost.
The question behind the search
Nobody searches "Bob vs Codex" because they are choosing who writes their code. They search it because the engineering team is having a great month with an AI agent, and everyone else is still building the Monday report by hand, copying numbers from the CRM into a spreadsheet, and chasing the same three updates every Friday. The fair question is whether the tool the engineers love can take that work too, or whether the rest of the team needs something else.
What Codex is
Codex is OpenAI's agent for coding tasks. According to OpenAI's pricing page it runs on the web, in the CLI, in the IDE extension, and on iOS, and the cloud version lets an engineer start work from the web, GitHub, GitLab, Linear, or Slack, run tasks in parallel, then review the summary and diff and open a pull request when it is ready. OpenAI's plan descriptions are written for developers: Plus is for "a few focused coding sessions each week".
Codex does have a Slack app. You mention @Codex in a channel or thread, and OpenAI says it creates a cloud chat and replies with the results. To set it up, OpenAI's page lists a Plus, Pro, Business, Enterprise, or Edu plan, a connected GitHub account, and at least one environment, meaning a code repository Codex can work in. That is a sensible design for engineers, and it tells you who the Slack app is for.
It is also careful. Locally, OpenAI describes a sandbox plus an approval policy that decides when Codex must stop and ask, and it says destructive app tool calls always require approval when the tool marks itself as destructive.
What about the rest of OpenAI?
Here is the part that surprised us, and it is worth knowing before you decide. OpenAI now offers ChatGPT Work, "a way to delegate real work to ChatGPT": briefs, decks, analyses, recurring updates, and files you can review. It runs on the Codex harness, it shares usage with Codex, and OpenAI's own guide says that if you have used Codex for non-coding work, you can stay in Codex or use ChatGPT Work instead. It can keep running in the cloud after you close the app, and scheduled tasks can start from new Gmail, Slack, or GitHub events on eligible plans.
So the honest answer is that OpenAI does have a product for non-coding work, and it is a serious one. It lives inside ChatGPT, where a teammate delegates a task, follows progress, and approves important actions, and on ChatGPT Business it is included in a per-user plan. If that is the shape you want, it is worth a look.
What Bob is
Bob is not a better chat window. Bob is a hire.
Bob does the work that wastes your day. He lives in your team's Slack, so anyone can hand him a job the way they would a new colleague: @mention him in the channel where the work is already being discussed. He signs into the apps your company runs on, 3,200+ of them, and personal ones connect per person, so a request runs under your access and not a shared login's. He opens the spreadsheet, pulls the numbers from the CRM, builds the deck, and posts the finished thing back in the thread.
He has his own computer in the cloud, so the Monday rollup is posted before you open Slack, whether or not your laptop is on. Before a sensitive write he stops and asks, in the thread, with Approve, Always allow, and Deny, so the whole channel sees who said yes. Every finished run ends with a receipt: what it cost, how long it took, and what he touched. He works in Slack today, with Microsoft Teams coming soon.
Side by side
Each row is a question a buyer asks anyway. The Codex column says what OpenAI's pages said on September 23, 2026, with a link to each one, and where those pages do not cover something, this table says so rather than guessing.
| HeyBob | Codex | |
|---|---|---|
| Who is it for? | The whole team. Ops, sales, finance, support, and founders hand him jobs in Slack. | Developers. OpenAI describes Codex cloud as running coding tasks and points everyday non-coding work to ChatGPT Work. |
| Where does the work happen? | In your Slack threads and your connected apps. Microsoft Teams is coming soon. | The web, CLI, IDE extension, and iOS, with cloud tasks started from GitHub, GitLab, Linear, or Slack. |
| What does it take to start in Slack? | Add Bob to Slack, then say "connect HubSpot" and sign in once. About five minutes to the first finished job. | A paid ChatGPT plan, a connected GitHub account, and at least one environment. |
| What comes back? | The finished thing in the thread: the answer, the workbook, the deck, or the pull request. | In Slack, the result and a link to the cloud chat. In the cloud, a summary and diff to review before opening a pull request. |
| What happens before it writes? | Sensitive external writes pause in the thread with the app and action named, plus Approve, Always allow, and Deny. The decision stays visible to the channel. | A sandbox and approval policy the person running it configures. Destructive app tool calls that are marked as destructive always ask. |
| What can you see afterwards? | A receipt under every finished run: credits, time, tool calls, and who approved what. Ask "what ran this week" and he adds it up. | The task summary and diff. Enterprise adds audit logs and usage monitoring via the Compliance API. |
| How do you pay? | From $75/month for the whole workspace, no seats. Hard monthly cap, unused credits roll over a month, free reruns on his mistakes. Or $349/month flat on your own OpenAI or Anthropic key. | Included in ChatGPT plans from Free to Enterprise. Business is $20 per user per month billed annually, $25 monthly, with credits past the included usage. |
What the rest of the team hands Bob
The easiest way to see the difference is the work itself. None of this touches a code repository, and all of it eats hours every week.
| The job | What Bob does |
|---|---|
| The Monday report | Pulls the numbers from the CRM and the spreadsheet, writes the summary, and posts it in the channel before anyone asks. Every week, on its own. |
| Copying between apps | Moves the deal notes into the tracker, the invoice into the books, the ticket into the project board, and asks before anything goes out. |
| "What do our numbers say?" | Opens the file, does the math in it, and answers in the thread, with the workbook attached. |
| The customer update | Drafts it from the account history and waits for your Approve before it sends. |
And when the job does touch code, like finding a bug someone reported in Slack and opening the fix, Bob can do that too. Your engineers will probably still want Codex in their own editor, and that is fine.
The part nobody puts in a comparison table
These two are not really rivals. OpenAI builds some of the best models in the world, and Bob can run on them: on the Bring your own keys plan, your OpenAI key powers Bob and the model bill goes straight to OpenAI. The difference is the shape of the thing. Codex is an agent your engineers drive from their own tools. Bob is a teammate the whole company shares, in the place the company already talks, with an approval before anything sensitive and a receipt after everything. Plenty of teams will end up with both.
Keep Codex for your engineers. It is built for code, it lives where developers work, and OpenAI's own pages show how seriously they take its sandbox and approvals. If people on the rest of the team want to delegate their own tasks inside ChatGPT, look at ChatGPT Work.
Add HeyBob for everyone else: one AI employee the whole team shares in Slack, 3,200+ app connections under each person's own access, jobs that run on a schedule whether your laptop is open or not, an approval in the thread before sensitive writes, a receipt on every run, and one workspace price with a hard cap.
Fair questions
Our engineers already use Codex. Does it cover the rest of the team?
OpenAI describes Codex in Slack as a way to kick off coding work, and its Slack setup asks for a connected GitHub account and a code environment. For everyday non-coding work, OpenAI now offers ChatGPT Work, which lives inside ChatGPT. Bob is one AI employee the whole team shares in Slack, on one workspace price, with approvals and a receipt on every run.
Can we keep Codex and add Bob?
Yes, and that is the setup we expect most. Your engineers keep Codex for their code. Bob takes the reports, updates, and copying between apps that the rest of the team does every week, in the Slack channels where they already work.
Can Bob run on our OpenAI account?
Yes. The Bring your own keys plan is $349/month flat: paste your OpenAI or Anthropic key into settings and Bob's runs execute on it, so the model bill goes straight to your provider at your rates.
How is Bob priced compared with Codex?
OpenAI includes Codex in ChatGPT plans, with ChatGPT Business listed at $20 per user per month billed annually, or $25 billed monthly. Bob has no seats: plans start at $75/month for the whole workspace, every finished run shows what it cost, you can set a hard monthly cap, and the first $100 of work is free with no card.
Sources
Every OpenAI page below was checked on September 23, 2026.