OpenAI “Intelligence at Work”: How Codex and ChatGPT’s New Business Features Change the Way Companies Work (2026)

OpenAI’s “Intelligence at Work” is the moment Codex and ChatGPT stop being clever chatboxes and start being co-workers. The pitch is simple and consequential: instead of an assistant you visit in a separate tab, you get role-aware agents that sit inside the tools your teams already live in — the CRM, the BI stack, the design suite, the spreadsheet — and do the work, not just describe it. For B2B leaders, the headline isn’t a benchmark. It’s that a large slice of the manual work currently done by hand across sales, marketing, analytics, product and finance can now be handed to an agent that understands both your company context and your app stack. This guide unpacks exactly what shipped, what it costs, how to roll it out, and the ready-to-paste prompts we’re already deploying with clients.

By Toni Dos Santos, Co-Founder, Spicy Advisory — we help mid-market and enterprise teams actually use the AI tools they’ve bought, tool-agnostically, across the UK and EU.

What is OpenAI’s “Intelligence at Work”?

“Intelligence at Work” is OpenAI’s push to turn Codex and ChatGPT into an “intelligence layer” that runs across business systems rather than a standalone chatbot. Powered by the GPT‑5.x family and Codex as an agentic platform, the strategy reframes ChatGPT as an enterprise console — models, tools, connectors, agents and analytics in one place — while Codex becomes the tool-native agent that builds, operates and automates workflows across your apps and devices. Four launches make that concrete: six role-specific Codex plugins, Codex Sites, Annotations, and Codex moving inside the ChatGPT app everywhere. Together they signal that Codex is no longer “the developer tool” — it’s a general productivity surface non-technical teams can drive in plain language.

What’s new at a glance

Here is the whole release in one table, framed for a business reader deciding where to start.

What shipped	What it does	Who it’s for
Six role-specific Codex plugins	Bundle the apps, skills and workflows for a job into one agent — 62 apps and 110 skills across all six	Sales, marketing/creative, analytics, product design, public-equity investors, investment bankers
Codex Sites (preview)	Generate interactive, hosted internal web apps and dashboards from a prompt, shareable by URL	Any team that needs a dashboard, planner or review hub without waiting on engineering
Annotations	Select any element — a chart, paragraph or claim — and ask Codex to change it or show where it came from	Anyone refining documents, decks, spreadsheets and Sites
Codex inside ChatGPT	Codex’s agentic capabilities become available inside the ChatGPT app “everywhere” — one front door	Companies already standardised on ChatGPT who don’t want two tools

OpenAI has also signalled more roles on the way — corporate finance, private equity, marketing strategy, strategy consulting and legal — plus a wider partner ecosystem in both Codex and ChatGPT.

The six role-specific Codex plugins (and how each team uses them)

Think of each plugin as an off-the-shelf AI co-worker that already knows the workflows and the app stack for a role. They work out of the box and can be adapted. Below: what each one connects to, the business impact, and a ready-to-paste prompt you can try in a pilot.

1. Data Analytics plugin

Connects Codex to platforms like Snowflake, Databricks Genie, Hex and Tableau so business users can interrogate product and business data in plain language. The point is self-service BI: product managers and operators stop filing tickets to the data team for every question.

Prompt to try: “Using our Snowflake warehouse, find out why weekly active users dropped in France last week. Break it down by acquisition channel and device, generate the three charts that best explain the move, and write a two-paragraph summary I can paste into Slack.”

2. Creative Production plugin

Connects to Figma, Canva, Shutterstock, Picsart and Fal to take marketing and creative teams from brief to reviewable assets. Codex turns a campaign brief into campaign boards, ad sets, social variations and product lifestyle shots within brand constraints — then iterates on feedback instead of starting from scratch.

Prompt to try: “Here’s our Q3 campaign brief and brand guidelines. Produce a campaign board, four ad-set variations sized for LinkedIn, Instagram and Meta, and three product lifestyle concepts. Keep to our palette and tone, and lay them out in Figma for review.”

3. Sales plugin

Integrates CRMs and sales tooling — Salesforce, HubSpot, Slack, Outreach, Clay, Rox and Actively — to prioritise accounts, prep meetings, draft recaps and follow-ups, and keep pipeline hygiene tidy (notes, tasks and fields updated automatically). Deal-review flows surface at-risk opportunities and suggest the next action inside the tools reps already use.

Prompt to try: “Rank my open opportunities by risk using engagement and CRM signals. For the top five at-risk deals, summarise what’s stalled, draft a re-engagement email per account, and create the follow-up tasks in HubSpot.”

4. Product Design plugin

Built around Figma and Canva to turn early ideas into prototypes. Teams explore product directions and user flows, convert static screenshots into interactive prototypes, and even prototype from a live URL — with the work staying editable in the design tools.

Prompt to try: “Turn these three onboarding screenshots into a clickable prototype, then propose two alternative flows that cut the steps to first value. Keep everything editable in Figma so the team can react.”

5. Public Equity Investing plugin

Aggregates institutional data from Moody’s, Daloopa, Datasite, FactSet, LSEG, S&P, PitchBook and Hebbia so public-markets investors can review earnings, compare companies, monitor signals and test whether a thesis is strengthening or weakening — with Codex synthesising the data into structured views.

Prompt to try: “Review the latest earnings for these five names, compare margin trends and guidance against consensus, flag anything that changes our thesis, and produce a one-page memo with the evidence linked.”

6. Investment Banking plugin

Helps bankers turn research and diligence into client-ready materials from trusted financial datasets — analysing comparable companies and transactions, structuring pitch books, and translating dense diligence into decks, summaries and recommendations.

Prompt to try: “Build a comparable-companies analysis for this target, pull the most relevant precedent transactions, and draft the first eight slides of a pitch book with a clear valuation summary and key risks.”

Codex Sites: your internal app factory

Sites lets Codex generate interactive, hosted internal web apps you can share by URL inside your workspace — currently in preview for Business and Enterprise. Instead of waiting weeks for engineering to build a dashboard or tracker, a team asks Codex to assemble one from its existing tools and models in hours. Useful patterns:

• Customer review hubs that pull product updates, open questions, usage trends and next steps for a key account into one live page.
• Scenario planners built from a financial model, where assumptions are adjustable in the UI rather than buried in spreadsheet tabs.
• Launch hubs that keep messaging, milestones, owners and decisions current as Codex updates them over time.

OpenAI is working with partners including Vercel, Wix, Base44, Replit, Lovable, Figma, Webflow and Emergent, hinting that Sites may eventually bridge from internal tools to production-grade web apps. The trade-off to manage: when anyone can spin up a dashboard, you need standards for data sources, naming and ownership so the outputs stay trustworthy. (If you’ve seen our guide to building live artifact dashboards in Claude, this is the same idea on the OpenAI side.)

Annotations: in-place refinement and source traceability

Annotations extend Codex’s refinement from code and websites to business content — documents, spreadsheets, slides and Sites. You select a specific element (a nav bar, a chart, a paragraph, a claim) and ask Codex to change just that, without regenerating the whole document. Two things make this matter for business:

• Finer-grained editing. “Update this chart label for clarity” or “explain this metric in plainer language” — surgical, not wholesale.
• Traceability. Highlight a claim in an investment thesis or board paper and ask Codex where it came from, surfacing the underlying data or reasoning. For regulated and finance teams, that audit trail is the difference between “interesting” and “deployable.”

The pattern it reinforces is healthy: Codex drafts, humans refine against specific parts of the work, and nothing gets redone from zero each round.

Codex inside ChatGPT, everywhere

OpenAI says Codex functionality is moving directly inside the ChatGPT app “everywhere” in the coming weeks, so companies standardised on ChatGPT don’t have to consciously switch tools to use Codex agents. This resolves the most common question we get — “when do I use ChatGPT vs Codex?” — by letting people stay in ChatGPT while Codex-style agent capability runs behind the scenes. It builds on existing ChatGPT Business features: workspace agents that run scheduled workflows, an admin console with agent analytics, and deep integrations into Excel, Google Sheets, Outlook, Box, Notion and Google Drive. The strategic read: Codex is becoming the automation engine embedded in ChatGPT-based work, not a separate pro tool. If your company is standardised on ChatGPT, our ChatGPT enterprise training gets non-technical teams running these agents safely and consistently. (For the Microsoft-stack angle, see our ChatGPT × Microsoft Office integration guide.)

Pricing: what it costs in 2026

The role plugins, Sites and Codex-in-ChatGPT ride on existing ChatGPT and Codex plans rather than a separate SKU. Here’s the practical map as of June 2026 — always confirm current terms with OpenAI before procurement.

Plan	Indicative price	What you get for “Intelligence at Work”
Free	£0	Core ChatGPT; limited access to the latest models and agentic features — fine for evaluation, not for rollout
Plus	$20 / user / mo	Entry point for Codex and the agentic surface; good for individual pilots and power users
Pro	$200 / user / mo	Highest usage limits and heaviest agent/Codex workloads; for power users running many parallel agents
Business	~$25 / user / mo (annual), ~$30 monthly	Team workspace, admin console & analytics, connectors, workspace agents — and Sites in preview; the realistic starting tier for most companies
Enterprise	Custom	SSO, data residency, advanced admin/governance, audit logs, scaled seats and support; Sites preview included

How to budget it honestly. The licence is rarely the real cost — enablement and governance are. For a 50-person pilot on Business you’re looking at roughly $15k–$18k/year in licences; the bigger line items are the time to wire up connectors safely, train the people who will actually run the prompts, and review outputs. Start with one or two functions, prove ROI, then scale seats. Note that Sites is preview-gated to Business and Enterprise, and the deepest governance controls live on Enterprise — which matters if you’re in a regulated industry.

How to roll this out: a step-by-step guide

This is the sequence we use with clients to get from announcement to measurable value without creating shadow IT.

1. Pick one function and one painful workflow. Not “roll out AI” — pick, say, “sales deal-review prep” or “weekly product KPI root-cause.” A narrow target gives you a clean ROI hypothesis.
2. Map the app stack and the data it touches. List exactly which systems the matching plugin will read from and write to (e.g. Salesforce, Slack, Snowflake). This is your security and connector checklist before anything is enabled.
3. Stand up a Business or Enterprise workspace with the right plugin. Enable only the connectors the pilot needs; restrict write access at first so agents draft and humans approve.
4. Write 3–5 “golden prompts.” Turn the workflow into reusable prompts (start from the examples in this guide). These become your team’s playbook, not one-off cleverness.
5. Run a two-week supervised pilot. Humans review every output. Track time saved, quality, and where the agent gets it wrong — that error log is gold for training and for setting guardrails.
6. Decide attended vs unattended per task. Nightly report generation can run unattended; client communications and investment recommendations stay human-in-the-loop. Write the policy down.
7. Add a Site or scheduled agent once trust is earned. Convert the proven workflow into a Codex Site (a live dashboard or review hub) or a scheduled agent, then measure adoption in the admin analytics.
8. Govern, then scale. Review connector scopes, set up audit logging, name an owner, and only then expand to the next function. Treat AI as an internal platform with templates and standards — not a gadget.

Business impact by function

Where the value actually lands, with concrete moves for each team.

Sales & Customer Success

Use the sales plugin plus Sites to automate account planning, QBRs and follow-ups. Maintain a live Site per strategic account with health metrics, open risks and renewal plans; auto-generate QBR decks from CRM and product-usage data; let agents keep pipeline fields and tasks current so reps sell instead of admin. (See our AI training for sales teams.)

Marketing & Creative

Orchestrate end-to-end campaign workflows: strategy in ChatGPT, asset production via the creative plugin, performance dashboards via Sites. The win is fewer handoffs between the planning doc, the design tool and the analytics tab — and far faster A/B experimentation on messaging and visuals. (See our AI training for marketing teams.)

Product, Analytics & Operations

Combine the analytics and product-design plugins for a closed loop: run “why” analyses on product KPIs in plain language, turn the result into a dashboard or Site for stakeholders, prototype the fix in Figma, validate, and hand engineering a clear brief. Ops teams can schedule agents for inbox triage and recurring reporting. (See our AI training for data & analytics and for product teams.)

Finance & Investment

Public-markets and advisory teams can run always-on investment and deal screens that refresh with new data, generate evidence-backed memos, and turn research and diligence into structured, annotated Sites where every claim is traceable to its source. With the upcoming corporate-finance plugin, FP&A teams get automated reporting and scenario tools. (See our AI training for financial services.)

Leadership & Strategy

Build live steering dashboards as Sites that combine metrics, qualitative updates and a decision log across Slack, BI, CRM and docs — so the weekly exec review assembles itself from working tools instead of a Sunday-night slide marathon. Spin up scenario-planning Sites for headcount, revenue and product bets. (See our C-level AI training.)

Where this sits next to Claude and Copilot

If your team already chose ChatGPT, the honest read is that “Intelligence at Work” closes most of the workflow-automation gap that had opened in Claude’s favour over the last year — without forcing a vendor swap. The pragmatic 2026 posture for most enterprises is still two vendors by use case, not one: ChatGPT + Codex for day-to-day role workflows and internal apps; Claude for long-context document reasoning and coding-heavy work; Microsoft Copilot where the Microsoft 365 graph is the centre of gravity. We dig into that decision in our Claude vs ChatGPT for business guide, and this release is the natural successor to the April 2026 Codex update we covered earlier. The point isn’t to pick a winner — it’s to give employees one front door and let IT expose the right capability behind it.