AI・テクノロジー
ChatGPT Full Automation | The Shocking Full Picture of Workspace Agents
機械翻訳 / Machine-translated
A
AI Friends
2026年4月26日・28分で読める
機械翻訳 / Machine-translated
@aifriends
AI Friends（https://aifriends.jp）のクロスポスト公式アカウント。AIツールの紹介・使い方・できることを、中学生でもわかるやさしい日本語で届けます。
"When will the day come that I can hand off all my work to ChatGPT?" — That dream has finally taken shape.
On April 22, 2026, OpenAI unveiled "workspace agents," evolving ChatGPT from a "smart advisor" into a "colleague who actually gets things done."
Here's a thorough breakdown of what's changing and what it means — including the impact of SoftBank's adoption.
First, let's get clear on what this announcement actually brings.
Workspace agents are the evolved successor to the existing "GPTs."
If GPTs were "robots you build as a personal hobby," workspace agents are "full-time employee robots that the whole company can use" — that's the mental image.
The defining feature: a single agent can be developed by the team and called upon by anyone through ChatGPT or Slack.
The old GPTs will remain available for now, but OpenAI has explicitly stated it will "progressively make migration to workspace agents easier," making this the main player for enterprise use.
Under the hood is OpenAI's coding-specialized model, "Codex." While ChatGPT itself only responded with text, Codex goes further — "actually writing and executing code," "editing spreadsheets," and "pressing buttons in external apps."
Think of it as the difference between "a receptionist" and "an accountant who processes invoices." Since it can run continuously in the cloud for extended periods, it can even be used for overnight batch-style processing.
Traditional ChatGPT operated on a "one question, one answer" basis.
Workspace agents handle multi-step workflows.
For example: "When a customer inquiry arrives → search past history → update Salesforce → notify the person in charge via Slack → draft a follow-up email" — a five-step chain of tasks carried out from start to finish without humans needing to prompt it repeatedly along the way.
Think of it like "a relay runner who runs all the way to the finish line without handing off the baton."
Here are the use cases highlighted in OpenAI's official announcement, adapted to the Japanese business context.
The "Sales Opportunity agent" automatically researches the latest information on clients → summarizes Gong call recordings → posts deal briefs to Slack's sales channel.
It's like arriving at your desk to find materials for the five companies you're visiting tomorrow already laid out before your first morning meeting — drastically cutting prep time for sales reps.
The "Lead Outreach agent" goes even further: researching inquiry content → scoring leads → drafting reply emails → updating the CRM, all in one seamless flow. It has the potential to double or triple the number of deals each salesperson can manage simultaneously.
For product teams, the "Product Feedback Router" is invaluable. It continuously monitors Slack, support chats, and public forums (X and Reddit), performs sentiment analysis on user complaints, and converts high-priority requests into tickets.
It's equivalent to "hiring a dedicated staff member who stays awake 24 hours a day listening to the customer's voice."
The "Weekly Metrics Reporter" automatically aggregates weekly KPIs → generates charts → shares a report complete with a summary. Work that used to take a responsible person half a day every Monday morning gets done in 15 minutes.
The impact on finance is particularly striking.
The "Month-end close agent" is said to complete — in a matter of minutes — journal entry creation → balance sheet reconciliation → variance analysis → generation of workpapers with source data attached.
"The era when accounting staff who used to pull all-nighters at month-end can actually leave on time" may genuinely be arriving. For IT departments, an agent that automatically reviews employee software requests → checks them against security policies → automatically issues tickets has also been introduced.
Some may worry, "AI acting on its own sounds scary." OpenAI has prepared robust controls for administrators.
Administrators can configure in detail — by role — "who can create agents," "who can run them," and "which tools they can connect to." Additionally, approval flows can be set up for more consequential actions like "sending email," "adding calendar entries," and "editing spreadsheets."
It's the same thinking as "not letting a new employee submit documents without a stamp of approval at first." There's also a switch that lets administrators stop any suspicious activity in real time.
Administrators can pre-whitelist which apps and data sources agents are allowed to connect to.
Restrictions like "this agent only has access to Slack and this specific Google Drive folder" are possible.
It's a design that structurally eliminates the risk of "AI secretly sending files to servers outside the company." A feature to visualize the operating history and connected data of all agents from the admin console has also been announced as coming soon.
On pricing: it's a generous offer of completely free, unlimited use until May 6, 2026.
Available plans are ChatGPT Business ($20/user/month, approximately ¥3,000), Enterprise, Edu, and Teachers — four plans in total. It is not available on the personal Free, Plus, or Pro plans.
This is structured as "a feature you can only access with a company contract." Enterprise pricing is customized based on headcount and features, with dedicated credit pools available for large corporations.
After May 6, the system shifts to a credit-based model.
Think of it as "the company buys a prepaid Suica card, and each time an agent does something, it's deducted."
The Business version includes a set amount per user in the flat rate, with additional credits purchasable for overages. The Enterprise version has the entire workspace sharing one large pool of credits.
Billing is token-based, meaning credit consumption varies depending on the model and features used (code execution, image generation, etc.).
The enterprise AI agent market is fiercely competitive. Let's compare against the main rivals.
"Microsoft stays inside Excel, Anthropic is a serious craftsman, Google is strong within its own office, OpenAI is the jack-of-all-trades" — a picture of distinct niches is emerging. For Japanese companies outside the Microsoft or Google ecosystem, workspace agents are likely to be the top candidate.
The biggest thing Japanese business professionals should pay attention to is that SoftBank is listed among the early adopters.
OpenAI's official announcement lists Rippling (HR tech), SoftBank Corp., Better Mortgage (home loans), BBVA (a major Spanish bank), and Hibob (HR platform) as the five early testers. SoftBank in particular, with Chairman Masayoshi Son pushing an "AI × Telecom" strategy, is the domestic company advancing AI-driven internal operations at the fastest pace.
The ripple effect — "if SoftBank is using it seriously, other large companies will follow" — is a certainty.
ChatGPT Enterprise and Business already operate fully in Japanese, and workspace agents are also expected to be designable and operable with Japanese-language instructions.
Anticipated domestic use cases include "an agent that shares a weekly sales report on Slack every Monday morning" and "an agent that automatically categorizes customer emails and routes them to the appropriate team." The benefits of adoption are likely to be especially significant in the accounting, general affairs, and customer support functions of small and medium-sized enterprises struggling with labor shortages.
There are caveats to note, however. Because Japanese companies have complex approval flows — the "hanko culture" of official stamps, the "ringi culture" of formal sign-off processes — the key to successful implementation is figuring out how to integrate workspace agents' approval features into Japan's decision-making routes.
To prevent accidents where "AI places orders on its own," the standard approach is to start small with read-only and aggregation-type agents. Systems integrators (SIers) and other implementation support partners are expected to announce their services in rapid succession going forward.
Yamada-san, editor-in-chief at a startup media company, starts his morning differently now.
Overnight, a "news monitoring agent" has browsed 30 sites and already posted summaries of 10 new articles on AI, Web3, and startups to Slack.
A "competitive analysis agent" has analyzed PV trends at rival outlets and submitted a hypothesis memo noting "this topic looks like it'll perform well this week."
"Work that used to take two researchers half a day to prepare for an editorial meeting now gets done in the time it takes to drink a cup of coffee" — Yamada-san can now focus on what he does best: coming up with ideas.
Tanaka-san handles all accounting solo at an 80-person machinery parts manufacturer.
Month-end always meant endless overtime.
Everything changed after she built a "monthly closing agent" with workspace agents.
The AI now handles reading invoice PDFs → creating journal entries → entering data into accounting software → checking for variances → creating a summary for the president.
Tanaka-san can now focus exclusively on the most critical task: verifying that the numbers make sense.
Month-end overtime dropped from 40 hours to 5 hours, and she got her time with family back.
Suzuki-sensei handles academic affairs at a middle school, using workspace agents through the ChatGPT Teachers plan.
An "parent communication agent" automatically sorts absence-notification emails and notifies homeroom teachers via Teams.
A "test grading assistance agent" performs a first-pass check on written answers, and a "grade aggregation agent" analyzes trends across the entire grade level.
"The administrative work after school has decreased, and there's more time to engage with students" — the mood in the staff room has brightened.
A. Not at this time. It is not available on Free, Plus, or Pro plans, and requires a corporate contract at Business ($20/person/month) or above.
The most practical approach is to verify it through a short-term contract like "try Business for just one month." Since Business can be started with as few as one user, sole proprietors can also sign up.
A. GPTs will continue to work as-is for the time being. OpenAI has explicitly stated it has "no plans to discontinue GPTs," and a feature to "convert GPTs into workspace agents" is coming soon.
It's being treated like the period when feature phones transitioned to smartphones — there's no need to rush the switch.
A. ChatGPT Business and Enterprise contracts specify that input data will not be used for OpenAI's training.
Furthermore, administrators can restrict "which tools and data can be connected," which structurally suppresses the risk of "confidential data being secretly sent externally."
For particularly sensitive operations such as finance or healthcare, the safe approach is to start with non-confidential tasks and expand into sensitive areas once you're comfortable.
A. No. Just click "Agents" in the ChatGPT sidebar and describe the workflow you want in natural language.
Say something like "Every Monday, summarize the posts in this Slack channel and save them to this shared drive," and ChatGPT will organize the steps and build the agent for you.
It's an era where you can design AI the same way you'd "explain a recipe to a friend."
A. OpenAI has already established a Japanese subsidiary, OpenAI Japan LLC, and Japanese-language support is provided for Enterprise customers. Business-tier support is primarily in English via the official help center, but Japanese SaaS resellers and AI consulting firms are expected to launch implementation support services in rapid succession.
The option to "rely on a domestic partner when you hit a wall with implementation" is taking shape.
We are witnessing the moment AI evolves from "just answering" to "actually doing."
"Month-end overtime disappears," "sales reps can handle twice the caseload," "teachers get more time with students" — the changes brought by workspace agents may go beyond mere efficiency gains, posing a fundamental question to all of us: "What do we work for?"
Don't fear it. Start small, and get comfortable.
That's the trick to thriving in the age of AI agents.
This article is a cross-post from AI Friends.
