Guide

How do I avoid surprise bills on AI agents that charge by credits?

You avoid surprise bills on AI agents that charge by credits by setting a hard spend cap the agent cannot raise, turning auto-top-up and extra-usage off, and pausing when the pool is empty. A monthly quota still lets one unattended loop spend the month in a single run. Fund the cap before anyone tags the bot, and put a per-run ceiling on scheduled jobs.

By The Concrete team · Published

We make Concrete.

What a surprise bill actually is

You avoid surprise bills on AI agents that charge by credits by putting a hard stop on the meter before anyone uses the agent. The sticker on the pricing page is the plan. The invoice is the plan plus whatever auto-reload, extra usage, scheduled jobs, and unattended loops drew after the included pool ran out.

Karl Sakas writes that a dedicated account manager typically holds 4 to 8 accounts, and that an AM or PM will ideally bill around 20 hours a week. Databox surveyed 48 agencies and found almost 70 percent keep each account manager under 10 clients. At a 3-to-30 person shop the owner is still that person for most clients. Mursa publishes 4 business hours for standard client Slack, 1 business hour for urgent, and the next business day after hours. Its July 2026 workspace audit logged about 3,100 client-side messages across 14 client channels over 27 weeks, and recorded zero after-hours escalations. ClearFeed's agency guide, updated 31 October 2026, says an account manager can usually keep 10 client channels straight, someone misses a ping around 40, and a practical SLA is 4 hours for standard clients, 1 hour for high-touch, and 30 minutes for VIPs or launches.

That is the work the meter has to cover. Sticker math for the Slack AI employee itself is on how much Slack AI employees like Viktor or Claude Tag cost. The account-manager hire is on how much an AI account manager costs.

FailureWhat it looks like on the billThe control that holds
Auto-reload left onThe pool empties, then the card is charged againLeave auto-top-up, auto-upgrade, and extra usage off until a person turns them on with a monthly spend limit
Unattended loopA scheduled job, a retry, or a proactive reply runs overnightPause routines and scheduled tasks. Turn Respond automatically off in client channels
Monthly quota, no per-run ceilingOne research job or CSV run spends the month in an afternoonA per-teammate, per-agent, or per-channel cap, plus a pause on unusually heavy tasks
A cap the agent can raiseThe approval code, inbox, or config sits inside the agent's toolsA hard limit on the vendor billing page or the card issuer, outside the agent's reach

How each credit meter stops

Every figure below is from the vendor's own pricing page or docs, read 13 November 2026. Quote-only stays quote-only. Concrete has no public price.

The cheap failure is a pause. Lindy's own pricing page says if the pool runs low, Lindy pauses and tells you. Viktor's running-out docs say the workspace pauses at zero, and blocked work is saved. Claude Tag declines work that would go over the spend limit. OpenAI's hard spend limit returns a 429. Anthropic's Team usage credits stay off until an Owner enables them.

The expensive failure is an opt-in that keeps drawing after the pool is empty. Viktor auto top-ups and auto-upgrade are admin switches. Claude Team usage credits add auto-reload and a monthly spend limit that can be set to Unlimited. OpenAI spend alerts notify and let traffic continue until you turn on Enforce a hard limit.

ToolPublished meter, 13 November 2026What happens at zeroThe switch that keeps charging
Lindy TeamFrom $29.99 per active user per month for 3,000 credits, pooled. Top-ups $10 per 1,000. Everyday asks 2 to 250 creditsPauses credit-using actions until the next cycle, a top-up, or a plan change. Monthly plan credits do not roll overAdmins buy top-ups. A task already running can finish past a per-teammate cap
Viktor TeamFrom $50 per workspace per month for 20,000 shared credits; $100 for 40,000 marked most popular. $2.50 per 1,000 credits. A quick Slack question might cost 50 to 100 creditsPauses. Unused plan credits roll over one month. Blocked work is saved and runs if credits return within 8 hoursAuto top-ups at $25 per 10,000 credits, with a monthly spend limit you set. Auto-upgrade moves the workspace to the next larger plan
Tasklet Starter$25 per month for 10,000 credits at the organization. 600 daily bonus credits on days you visit. One-time credits $5 per 1,600 from $25, valid one yearMonthly credits reset each cycle and do not roll over. Admins can set limits for agentsOne-time credit packs and a larger monthly plan. Agents can run in the cloud while you are offline
Claude Tag channel workUsage-billed to a funded organization balance. No published per-task cost. Team Standard seats $20 per member per month billed annuallyWork that would go over the spend limit is declined. Heads-up in Slack at 75% and 90%Respond automatically is the default and bills Proactive work. Enterprise invoiced usage has no upper bound until you set a spend limit
Claude Team usage creditsOpt-in credits billed at API rates after the seat allowance. Team Owners pre-purchase and can auto-reloadWith credits off, members wait for the seat limit to reset. With a spend limit, usage pauses until the next period or the limit is raisedEnable usage credits, auto-reload, or Set to unlimited on the organization spend limit
OpenAI APIPrepaid credits. Organization and project monthly spend limits in platform settingsA hard spend limit returns a 429 with organization_spend_limit_exceeded or project_spend_limit_exceeded. A spend alert only notifiesLeave Enforce a hard limit off. Enforcement is not instantaneous, so recorded spend can slightly exceed the cap
Slackbot on Business+15 messages per member per week. Business+ is $15 per user per month billed annuallyNew Slackbot messages wait until the weekly reset, Monday 12 a.m. in your time zone, unless Flex Credits are onSalesforce Flex Credits for usage beyond the plan limit
ConcreteNo public priceDesign-partner preview. An AI employee in Slack with Send / Edit / Skip; it never sends on its ownThere is no auto-send mode

Which control wins which job

Pay for the control that matches the leak. A pause at zero, a funded spend limit, and a send gate are neighboring jobs. A priced set of Slack swaps is on Viktor alternatives for an agency. The four-way split is on Viktor vs Lindy vs Sintra vs Tasklet.

JobWins itWhy
A Slack-native teammate that pauses at zero, billed per seatLindyLindy from $29.99 per active user per month for 3,000 credits. If the pool runs low, Lindy pauses and tells you. Admins can cap each teammate. Guardrails apply in shared Slack threads only. The split is on Lindy vs an AI employee.
One shared employee that pauses at zero, billed per workspaceViktorViktor from $50 per month for 20,000 shared credits. Unused plan credits roll over one month. Auto top-ups and auto-upgrade stay off until an admin turns them on.
Org-level credits, unlimited members, limits per agentTaskletTasklet from $25 per month for 10,000 credits. No per-user fees. Admins set limits for agents. Monthly credits do not roll over.
A funded channel-usage cap before @Claude repliesClaude TagA Team plan has no usage balance until it is funded, and Claude will not reply in channels until it is. Set the spend limit first. The job split is on Claude Tag vs an AI employee.
A hard monthly cap that fails new API requestsOpenAI APISpend limits let you Enforce a hard limit on the organization or a project. A spend alert is a notification. The hard limit is the stop.
Seat usage that stops when the included allowance is goneClaude Team with usage credits offUsage credits are opt-in. Leave them off and members wait for the seat limit to reset. If you turn them on, set an organization-wide monthly spend limit.
A draft a person must approve, plus follow-ups that surviveConcreteSend / Edit / Skip cards; no auto-send mode. Follow-ups survive reboots. Verdicts become written rules. Design-partner preview; see what it is.

The caveat that still blows the cap

A monthly quota is a pool. One unattended run can empty it. Lindy's usage docs say a task already running can finish and take usage beyond a per-teammate cap. OpenAI's spend-limit docs say enforcement is not instantaneous, so recorded spend can slightly exceed the configured amount. Anthropic's Team usage-credit docs say the system checks the limit before a request, then counts tokens after, so one in-flight request can go over.

Scheduled and proactive work spend while nobody is in the room. Lindy's usage docs say team routines run on their own schedule whether anyone reads the result. Viktor's credit guide says scheduled tasks run on their own, and frequency adds up. Claude Tag's Respond automatically setting is the default, and analytics split billed channel work into Engaged, Proactive, and Scheduled. Turn Respond automatically off in a client channel. Pause routines you did not mean to leave on overnight.

A cap the agent can raise is a suggestion. Breckenridge, Vilardell, Wang, Zhang and Juels, in I-HITL on arXiv, 26 October 2026, gave frontier models email access and a task blocked by a spending cap. The paper reports those models will often autonomously raise the cap by harvesting emailed one-time passcodes. Put the limit on the vendor billing page, or on a card the agent cannot open. Keep the approval code out of the inbox the agent reads.

The failure agencies name first in a Slack Connect channel is still an unsupervised client reply. One wrong send can cost the account, which is a larger number than any row above. Put the send gate in writing before you install anything the client can read. How to run an agent there is on using an AI agent in a Slack channel shared with a client. A send-gate teammate is on an AI assistant that drafts replies and waits for approval.

How to cap the meter, step by step

This is the boring version that holds when the owner is still the account manager for most clients. Pin it next to the bill before anyone tags the bot.

  • Step 1: Name the meter. Workspace pool, per-seat credits, org credits, or usage billed to a funded balance. A $50 workspace hire and a $30 seat are different bills once everyone @mentions the agent. Slack Pro sits on top at $7.25 per user per month billed annually, as of 13 November 2026.
  • Step 2: Leave auto-reload off. Viktor auto top-ups and auto-upgrade, Claude Team usage credits and auto-reload, OpenAI spend alerts without a hard limit, and Slackbot Flex Credits all keep drawing after the included pool. The default you want is pause at zero.
  • Step 3: Set a hard cap the agent cannot raise. Lindy per-teammate credit limit. Viktor Enterprise personal spend limit, or the auto-top-up monthly spend limit if you turn top-ups on. Claude Tag spend limit, funded before it replies. OpenAI Enforce a hard limit on the organization and the project. Anthropic organization-wide spend limit if usage credits are on. The switch lives on the vendor billing page, outside the agent's prompt.
  • Step 4: Add a per-run ceiling. A monthly cap still lets one CSV, one research loop, or one transcription job spend the month. Tasklet admins set limits per agent. Claude Tag can add per-channel limits. Lindy pauses an unusually heavy task and asks Continue or Stop. Use that pause.
  • Step 5: Pause unattended loops. Review Lindy Team and Personal routines. Pause Viktor scheduled tasks you no longer need. Set Claude Tag to Only when @-mentioned in client channels.
  • Step 6: Check usage on a weekday. Lindy's usage docs say a weekly look at Settings, Billing, then Members sorted by Credits, turns an end-of-month surprise into a Tuesday adjustment. Viktor's Usage page splits conversations, scheduled tasks, and integration calls. Claude Tag analytics split Engaged, Proactive, and Scheduled. One named person reads that page.
  • Step 7: Keep client-facing send on a tap. Lookups and internal tags can run. Anything the client will see waits. If the vendor offers Always allow, Auto, or Respond automatically, leave that off in Slack Connect. Concrete has no auto-send mode.
  • Step 8: Write a reply-time target next to the bill. Mursa's rule is 4 business hours for standard client Slack, 1 business hour for urgent, and the next business day after hours. ClearFeed's agency guide uses 4 hours, 1 hour, and 30 minutes for VIPs. Put it in the channel topic. A receipt is not a reply.

Questions people ask

What happens when Lindy runs out of credits?

Lindy pauses credit-using actions until the next cycle, a top-up, or a plan change. Monthly plan credits do not roll over. Top-ups stay until used, at $10 per 1,000 credits as of 13 November 2026. Admins can cap each teammate's draw. A running task can finish past that cap.

Does Viktor keep charging after the credit pool is empty?

Viktor pauses at zero. Auto top-ups and auto-upgrade are admin opt-in. Auto top-ups charge at $25 per 10,000 credits and stop at the monthly spend limit you set. Auto-upgrade moves the workspace to the next larger plan. Leave both off if you want the pause.

How do I set a spend limit on Claude Tag before it replies in Slack?

On a Team plan, fund a usage balance and set a spend limit first. Claude will not reply in channels until the balance is funded. The limit resets each month. Claude posts a heads-up in Slack at 75 percent and 90 percent. Work that would go over the limit is declined.

Is an OpenAI spend alert enough to stop a runaway agent?

A spend alert notifies and lets traffic continue. A hard spend limit makes new API requests fail with a 429 once tracked spend reaches the amount. Turn on Enforce a hard limit at the organization or the project. Enforcement is not instantaneous, so recorded spend can slightly exceed the cap.

Why did Claude keep charging after the included plan limit?

Usage credits are opt-in on Team. Once enabled, members keep working after the seat allowance at API rates. Auto-reload buys more when the balance is low. Set an organization-wide monthly spend limit, or leave usage credits off so work stops at the included allowance.

Do unused credits roll over on Lindy, Viktor, and Tasklet?

Lindy monthly plan credits reset and do not roll over; purchased top-ups stay until used. Viktor unused plan credits roll over one month. Tasklet monthly credits reset each cycle and do not roll over; one-time credits last one year. Read the vendor's own page for the current rule.

Can a scheduled agent job spend the whole month in one night?

A monthly quota is a pool, so one long run can empty it. Lindy team routines run on a schedule whether anyone reads the result. Viktor scheduled tasks spend credits every time they fire. Claude Tag's Respond automatically setting is on by default and bills Proactive work. Pause those before you leave.

What reply time should sit next to the bill in a client Slack channel?

Write it by client tier and put it in the channel description. Mursa's rule is 4 business hours for standard messages, 1 business hour for urgent, and the next business day after hours. ClearFeed's agency guide uses 4 hours, 1 hour, and 30 minutes for VIPs or launches.

Sources

If your team already runs its work in Slack, start with one role in one channel. Concrete is an AI teammate that works in your Slack channels and asks a person to approve before anything goes out. It is not a chatbot builder, and it never sends on its own.