Salesforce's $125 AI Problem: Why ChatGPT, Claude, Copilot and Grok Could Undermine Agentforce
Salesforce (NYSE: CRM) charges $125 per user per month for its Agentforce add-ons, while assistants that cost about $20 a seat can now read and update the same CRM records. Two of those integrations, for Claude and ChatGPT, were built by Salesforce itself.
That is the risk in one picture. Salesforce is opening its platform to every outside agent and betting it can charge for the access. If the bet fails, it keeps the database and loses the premium AI layer investors are counting on to replace slowing seat growth.
The market is already nervous. The stock fell 9.41% on November 17 to $194.62, its worst session since May 2024, in a sector-wide selloff of seat-based software with no company-specific news. That left it about 26.5% lower for 2026, two weeks before third-quarter results on December 2 (TradingNews, Salesforce IR).
The pricing problem
Agentforce is not one price. Salesforce sells it by the seat, by the action and by the conversation, and the options are not interchangeable.
Option | List price (USD) | What it covers |
|---|---|---|
Agentforce add-ons (Sales, Service, Field Service) | $125 per user per month | Unmetered employee use, on top of the underlying edition |
Agentforce Industries add-ons | $150 per user per month | Unmetered use plus industry AI |
Agentforce Coworker license | $50 per user per month | All Coworker searches and 30 MB of indexing per user per month |
Agentforce user license | $5 per user per month | Access only; usage draws Flex Credits |
Flex Credits | $500 per 100,000 credits | 20 credits, about $0.10, per standard action |
Conversations | $2 per conversation | Customer-facing agents |
Max editions | From $550 per user per month | Add-on plus a pool of Flex Credits |
List prices as reported from Salesforce's Agentforce pricing page by Reflekt Lab on November 4, 2026 and Macha.
For 100 users, the $125 add-on costs $150,000 a year before the underlying Salesforce licenses. At $0.10 an action, that seat price equals 1,250 metered actions per user per month, so the add-on only pays for heavy daily users.
Salesforce has already moved down-market. The $50 Coworker seat arrived on November 4, though its listed allowances cover searches and indexing and do not mention actions (Salesforce, Beri).
Four alternatives
All four major assistants can now reach Salesforce records, and three can write back out of the box.
Assistant | Route into Salesforce | Built by | What it can do | Seat price (USD) |
|---|---|---|---|---|
Salesforce in Claude plugin, in beta since September 15, 2026 | Salesforce, with Anthropic | 37 sales skills: account research, call prep, pipeline review, CRM updates under the seller's own permissions | Team standard seat: $20 per month billed annually, $25 monthly | |
Salesforce plugin, including the Agentforce Sales app | Salesforce | Read and write: account briefs, pipeline review, querying, write-back | Business standard seat: $20 per month billed annually, $25 monthly | |
Salesforce CRM connector; hosted MCP server through a Copilot Studio agent | Microsoft | Connector searches indexed records; record updates need the MCP agent route | Copilot Business: $21 per month list, $18 promotional through December 31, 2026 | |
Built-in Salesforce connector | xAI | Search records; create and update leads, opportunities, accounts and cases | Grok Business: $30 per month; Enterprise by quote |
Each comes with conditions. The Claude plugin is open only to organizations Salesforce approves through its beta sign-up. Copilot Business is limited to organizations of up to 300 users and needs a Microsoft 365 plan; larger firms pay $30 (Velosio). Seat prices are from Claude support, OpenAI help and TechRadar.
The detail that matters for investors is the "built by" column. Salesforce is not being worked around. It is placing its own data and skills inside rivals' interfaces, which concedes that many users will do CRM work somewhere other than Salesforce.
Other routes include n8n and custom agents connected to Salesforce APIs.
Why MCP matters
Model Context Protocol lets a compatible assistant call approved tools on an outside system: assistant, then Salesforce MCP server or API, then CRM records. Salesforce still controls authorization and the underlying data.
Salesforce's own hosted MCP servers come with Enterprise Edition and above, with no separate MCP product to buy (MindBlaze). The connection is free; what the tools do is not always free.
The bigger change is still pending. In a help article dated September 17, 2026, Salesforce said outside AI agents will have to be registered, and each successful call a registered agent makes, over MCP or the API, will count as a "Headless Platform Interaction" paid in Flex Credits (Orchestri, Vantage Point).
Three facts about that meter shape the investment case:
- The price is not public. The rate card lists the credit amount per interaction as to be announced.
- Metering has not started, and Salesforce has promised 30 days' notice before it does.
- Only production orgs are metered, and traditional integrations keep their current pricing.
So avoiding the Agentforce add-on does not mean avoiding Salesforce fees. It means swapping a known seat price for a usage fee nobody outside Salesforce can yet calculate. Open question: public material does not say whether Salesforce's own plugins inside Claude and ChatGPT fall under the meter.
What 100 users cost a year
On headline subscriptions alone, a general assistant is $114,000 to $126,000 a year cheaper than the $125 add-on for 100 users.
Option | Seat price (USD per month) | Annual cost, 100 users | Difference from $125 add-on |
|---|---|---|---|
Agentforce add-on | $125 | $150,000 | |
Agentforce Coworker | $50 | $60,000 | $90,000 lower |
Grok Business | $30 | $36,000 | $114,000 lower |
Copilot Business (list) | $21 | $25,200 | $124,800 lower |
Claude Team, standard seat, annual billing | $20 | $24,000 | $126,000 lower |
ChatGPT Business, standard seat, annual billing | $20 | $24,000 | $126,000 lower |
These are subscription comparisons, not total cost of ownership. They exclude the underlying CRM licenses, integration work, governance, support, and differences in what each product does. The assistant seats also carry usage limits, while the $125 add-on is unmetered.
The missing number is Salesforce's charge for outside agents. A $126,000 gap is $105 per user per month. If each metered call cost the same $0.10 as a standard Agentforce action, which is an assumption and not a published rate, the saving would vanish at about 1,050 calls per user per month, or roughly 50 per working day.
Below that volume the outside assistant wins. Above it, Salesforce recovers the revenue through usage. Agentforce's own consumption plans may also be cheaper than the add-on for light workloads.
The deeper bear thesis: three ways the model can break
The cheap-assistant problem sits on top of a business whose organic growth is already in single digits.
Second-quarter fiscal 2027 revenue (May to July 2026) was $11.3 billion, up 11%. Informatica, acquired in November 2025, supplied just over 4 points of that, leaving organic growth near 6% to 7%. Third-quarter guidance of $11.42 billion to $11.5 billion again includes slightly more than 4 points from Informatica (TradingNews, Salesforce).
Agentforce is growing fast from a small base. Its annual recurring revenue passed $1.5 billion in the second quarter, up more than 240%, against full-year revenue guidance of $46.1 billion to $46.4 billion. That is roughly 3% of revenue (Yahoo Finance). MLQ reported that Salesforce widened the metric that quarter to include Slackbot and Headless 360, which makes the growth rate harder to compare (MLQ).
Against that backdrop, three risks compound:
- Seat compression. Salesforce built its revenue on one license per human user. If agents do work that licensed employees used to do, customers need fewer seats, whoever supplies the agent.
- Losing the interface. If the agent is Claude, ChatGPT, Copilot or Grok, the customer's AI budget goes to that vendor. Salesforce keeps the database and loses the premium automation revenue meant to offset shrinking seats.
- A price anchor. Once buyers see CRM work done from a $20 or $30 seat, a $125 add-on is harder to defend for routine tasks. The new $50 Coworker tier can be read as an early response to that pressure.
Bank of America made the seat argument in May, rating the stock Underperform with a $160 price target (INDmoney).
Counterarguments
The bull case is that Salesforce is choosing this openness and intends to be paid for it.
- Salesforce owns the gate. Every outside assistant works under Salesforce permissions, sharing rules and audit controls, and the planned meter would turn third-party agent traffic into revenue.
- The integrations are Salesforce's own. It built the Claude and ChatGPT plugins, which keeps its skills and data in front of users wherever they work.
- The numbers have not cracked. Current remaining performance obligation rose 14% to $33.5 billion in the second quarter, and Agentforce and Data 360 annual recurring revenue reached nearly $3.9 billion (Salesforce).
- Production agents are hard. Running agents reliably at scale, with compliance and workflow logic, is harder than reading and updating records from a chat window.
- Much is priced in. The shares trade near 11.7 times the midpoint of fiscal 2027 non-GAAP earnings guidance, against a five-year average near 27 (TradingNews). Guggenheim upgraded the stock this summer, arguing the AI bear case was out of step with results (TIKR).
A low multiple cuts both ways. It limits the downside if the bears are wrong, and it shows how little confidence the market has that AI revenue will outrun seat losses.
What investors should monitor
The December 2 report is the next test, and six numbers will show which side is winning.
- Organic growth excluding Informatica. About 6% to 7% last quarter; guidance implies 7% to 8%.
- The outside-agent meter. The credit rate per Headless Platform Interaction, the start date, and how customers react.
- Agentforce ARR on a constant definition. Whether growth holds once newly included products are stripped out.
- Seat counts and net new bookings. Any sign that renewals come with fewer licenses.
- Mix between the $125 add-on and cheaper tiers. Uptake of the $50 Coworker seat and consumption plans at the add-on's expense.
- Production use and profitability. Paid Agentforce deployments in daily use, and the margin on consumption revenue.
Conclusion
Salesforce could win the CRM war and lose the AI interface war. Claude, ChatGPT, Copilot and Grok do not remove every reason to buy Agentforce, but they make a $125 add-on hard to justify for routine work.
The outcome now rests on one unpublished number: what Salesforce charges when someone else's agent touches its data. Set it too high and customers resist or route around it. Set it too low and Salesforce becomes indispensable infrastructure while other companies collect a growing share of the AI value.
This article is for information only and is not investment advice.
Sources
Primary pages cited in the original draft:
- Salesforce: Agentforce pricing
- Claude Help Center: Use Salesforce in Claude
- Salesforce Help: article 005467090
- OpenAI: Salesforce plugin
- xAI: Salesforce connector
- Salesforce Developers: Hosted MCP servers overview
- Salesforce: second quarter fiscal 2027 results and guidance
- Salesforce IR: third quarter earnings date
- Salesforce: Agentforce Coworker $50 option
- TradingNews: November 17 selloff, valuation and organic growth
- Orchestri and Vantage Point: metering of outside AI agents
- Yahoo Finance: Agentforce ARR
- Claude support, OpenAI help, Velosio, TechRadar: seat prices
- Microsoft Learn: Salesforce CRM connector and Vantage Point: Copilot through hosted MCP
This article is for general information only and is not financial, tax, legal or insurance advice. Check the official sources linked in the article and speak with a qualified professional about your own situation.

Comments
Post a Comment