Produce a defensible AI tool recommendation for a specific role or team. Return one recommendation, backed by the work people already do. A tool choice has four separate layers: durable foundation, daily chat, scheduled Cowork work, and full agent platforms. Do not let one layer decide the others.

The framework

Durable foundation

For APIs, internal tools, custom workflows, and agent work the company expects to keep, recommend Anthropic. This is the foundation choice. It does not settle the chat tab people use all day.

Use Claude Team for teams of 2 to 150 that need central billing, administration, SSO, and a predictable included-usage seat. Use Claude Enterprise when its governance controls are required: it is annual only, starts at 20 self-serve seats (50 through sales), and adds API-rate usage to the $20/seat monthly charge. It needs a usage owner and a spend cap.

Daily chat

For broad knowledge work, recommend the chat tool the role’s leveraged users already open. In most organizations, that is ChatGPT Business or Claude Team. ChatGPT Business is available from two users and includes centralized billing, administration, SAML SSO, analytics, budgets, and spend controls. Claude Team has the same basic buying shape for teams of 2 to 150.

Both standard seats cost $20 per user per month on annual billing or $25 month to month. If there is no existing preference, default to Claude Team. If the team has a real ChatGPT preference, recommend ChatGPT Business. ChatGPT Go is an $8 individual plan, useful only where company administration is not required. It is not a business-workspace substitute.

Do not recommend M365 Copilot as the only chat provision. It earns its place inside Word, Excel, Outlook, or a fixed suite-native workflow. Recommend it beside a baseline chat tool for doc-heavy roles where the team has proved that embedded workflow valuable. Use the customer’s actual Microsoft SKU and renewal price in the total, rather than guessing from a public list price.

Scheduled Cowork work

Recommend Cowork scheduled tasks for recurring work with a checkable output. A weekly competitive scan, a Monday status brief, or a reconciliation that flags what did not tie out. This is a narrower recommendation than a full agent platform.

Cloud Cowork is still beta, but remote work continues with the laptop closed and scheduled tasks can run with no device online. Work that needs local files or local apps still needs an awake machine. Local Cowork conversation history cannot be centrally managed or exported by admins, so make the location explicit: local or cloud. Every recurring task needs a named owner, an access boundary, a check, and a way to shut it down.

Full agent platforms

The serious vendor platforms are still immature. Recommend a time-boxed pilot, never a durable platform commitment. Choose two or three low-stakes workflows, give each a verifier, measure the actual failures, and revisit quarterly. The goal is to learn how the workflow behaves, not to declare a vendor winner.

Treat agent work as metered work even when it sits beside a flat seat. OpenAI Workspace Agents are token-priced, and Codex uses token-based credits. Before recommending either, require an owner, a usage forecast, a monthly spend cap, and an alert threshold. A fixed per-user total would be fiction.

Role tools

Choose the role tool from the work, not from the vendor relationship:

Engineering: Claude Code, Codex, Cursor, or GitHub Copilot. Let engineers choose from a real trial, then fund the tool they use.
Doc-heavy roles: M365 Copilot or Gemini where the work stays inside that suite. The value is the embedded workflow, not standalone chat.
Customer support: the AI built into the support platform. The ticket history and knowledge base are the point.
Sales: a CRM-native tool plus the baseline chat tool. A broad chat tool and a useful prompt library cover much of the work that stand-alone sales AI claims to sell.
Legal: a contract-review specialist once document volume justifies it, alongside the baseline chat tool.

Power users can have multiple subscriptions, API access, and the specialty tool they have earned. The annual-billing cost of both standard ChatGPT Business and Claude Team seats is $40 per person each month, or $480 per person each year. The month-to-month equivalent is $50 per month, or $600 per year. Do not force a power user through one interface when their work depends on both.

What you need from the user

Ask for six things:

Role or team. Who is this tool for? A specific job function (e.g., “marketing team of 12”) or a population (e.g., “all 200 knowledge workers”). The recommendation changes based on whether you’re buying for power users, the broad middle, or engineers.
Team size. How many people will have access.
What they do with AI today. Current tools, even if informal. Are people already using something? Are they paying out of pocket? Have they tried and bounced off something? This reveals what’s already working.
Work shape. Is this open-ended chat, a fixed suite workflow, durable internal tooling, recurring checkable work, or a proposed agent platform?
Administration. Central billing, SSO, SCIM, audit, data residency, or export requirements. This can change the plan even when the tool stays the same.
Spend shape. Budget ceiling, billing preference, expected agent or API volume, and the named owner of any metered work.

How to make the recommendation

Based on the role, work shape, existing use, administration needs, and spend shape, recommend one tool or one time-boxed pilot. Follow these rules:

One recommendation, not a shortlist. Name the tool, plan, and billing basis. A runner-up may appear only to explain the trade-off.

Keep the layers separate. Anthropic is the durable foundation default. Daily chat remains a role and preference decision. Cowork scheduled tasks are useful now for recurring, verifiable work. Full agent platforms stay in a pilot.

Default to what’s already working. If people on the team are already using a specific tool (even on personal accounts), recommend that tool. Adoption is harder than procurement. Don’t fight momentum.

Make administration decisive when it is. Move from ChatGPT Go to ChatGPT Business when central administration is required. Move from Claude Team to Enterprise when the Enterprise controls are required and the team can accept annual billing plus metered usage. Keep local Cowork out of a workflow that needs centrally managed or exportable history.

Put a meter beside agent work. Do not give a flat monthly total for Workspace Agents, Codex credits, API use, or Claude Enterprise usage. Show the seat total separately from the metered allowance, state the cap, and name the owner.

State when to revisit. Set the revisit date six months out. A platform pilot gets a 90-day decision date. Pricing, admin controls, and workflow evidence can change the recommendation sooner.

What to output

Produce a one-page brief with these sections:

Recommendation. The one tool or pilot, its layer, plan, billing basis, and why it fits this role.

Cost. Show the arithmetic. For a flat plan: seat count × monthly per-seat price = monthly total, then monthly total × 12 = annual total. For a metered plan: seat total + monthly usage cap = maximum monthly budget. Keep annual-billing and month-to-month numbers separate.

Administration and metering. State the controls that shaped the recommendation, the owner of metered work, and the usage cap or alert threshold.

Runner-up and trade-off. Name one runner-up only when it clarifies the choice. State what the team gives up.

Revisit. Give a six-month review date, or a 90-day decision date for a platform pilot. Name the evidence that could change the recommendation.

Don’t produce a comparison matrix. Don’t list more than two vendors.

Current plan and price sources

Anthropic pricing: Claude Team Standard is $20 per seat monthly on annual billing or $25 monthly; Team Premium is $100 or $125; Enterprise is $20 per seat plus usage at API rates. Team covers 2 to 150 people. Last verified August 14, 2026.
OpenAI Business pricing: ChatGPT Business is $20 per user monthly on annual billing or $25 monthly, from two users; Enterprise pricing is custom. Last verified August 14, 2026.
OpenAI’s ChatGPT Go announcement: ChatGPT Go is $8 per month in the U.S.; ChatGPT Pro is $200 per month. Last verified August 14, 2026.
OpenAI’s Codex rate card: Codex is token-priced in credits across applicable ChatGPT plans. Last verified August 14, 2026.
OpenAI’s ChatGPT rate card: Workspace Agents are token-priced and share an agentic usage and credit pool with related work features. Last verified August 14, 2026.
Anthropic’s Cowork release notes and Cowork enterprise guide: remote Cowork is beta, scheduled tasks can run with no device online, and local conversation history has separate administration limits. Last verified August 14, 2026.

Tool Selection Brief