Poke's AI Agent Is Easy to Use, Hard to Price

$29 per month. That is what one early Poke user reported negotiating the subscription down to, from a starting offer of $292, during the beta phase. The spread between those two numbers — a 10x delta on the same product — tells you everything about Poke's cost structure that the launch coverage did not. The monthly bill is not a fixed line item. It is a function of real-time inference volume. Poke's co-founders shipped the agent publicly in March 2026, and TechCrunch covered it on April 8. The coverage focused on the interface — AI agents accessed by text message, no app install, no onboarding flow. The cost mechanics received a fraction of a paragraph.

Whether Poke saves you money or quietly runs up a tab depends on one variable: the ratio of real-time inference requests to static queries in your usage pattern. A user who texts "remind me to call the dentist at 3 PM" consumes a sliver of the compute that a user running automated email triage on every incoming message requires. The difference is not marginal. It is structural. We will walk through three hypothetical users — composite illustrations, not real people — to show where the cost math lands for each profile.

Scenario 1: The Weekend Reminder User

Picture someone who picks up Poke for the simplest possible use case. Morning weather check. A grocery reminder. A calendar query before booking dinner. No automations. No integrations. No email processing. Five texts per day, maybe six on a busy Saturday.

Poke operates over iMessage, SMS, and Telegram — the user never installs a standalone app. This is the core premise that landed the startup a $15 million seed round from General Catalyst in 2025, at a $100 million pre-money valuation. The bet is straightforward: removing the app-installation barrier changes adoption curves for AI agents the same way WhatsApp changed them for messaging. By April 2026, an additional $10 million in funding had pushed the post-money valuation to $300 million. Previous valuation: $100 million at seed. Tripled in under a year. That velocity matters for pricing sustainability, which we will get to.

For this hypothetical user, the cost split is clean. Poke distinguishes internally between requests that require real-time data — live weather, flight status, current prices — and those that do not. The company's own framing, reported during the TechCrunch interview with co-founder Marvin von Hagen: "If you're asking for things that don't require real-time data, you could probably use Poke for free." A reminder does not need real-time inference. A weather query does.

Let us say this user sends five messages per day. Three are reminders or calendar queries — low compute cost, no live data pull. Two are real-time: weather and a transit check. Over a month, that is roughly 60 real-time inference calls and 90 static ones. The static calls are unlikely to push anyone off a free tier. The 60 real-time calls are the variable.

During beta, this usage profile sat comfortably in the $10–$15/month negotiation range — the lower end of what users reported after haggling with Poke's "bouncer" AI persona, which handled pricing during onboarding. The competitor benchmark is Lindy, an SMS-based personal assistant priced at $21/user/month, discounted to $18 through June 2026. Poke's paid tier starts at approximately $19.99/month. At this usage level, the two products are within $2 of each other per month. The differentiator is not price. It is which messaging platform you live in.

Monthly cost estimate for this persona: $0 to $19.99, depending on real-time query volume. The median outcome is probably $0 for the first month or two, $19.99 thereafter. Survivable.

Scenario 2: The Inbox Automator

Now imagine a different user. A consultant or solo operator who receives 80 to 120 emails per day and wants Poke to triage, classify, and draft responses. This is the use case that spawned Poke in the first place — von Hagen told TechCrunch that the company's original product was an AI assistant for email, and early testers started asking it to do everything else. "People wanted to use Poke for everything," he said. "They asked Poke to remind them to take their medication. They asked about sports results."

The cost structure shifts at this usage level. Every incoming email triggers a real-time inference call. Poke reads the email, classifies it, decides whether to surface it or handle it silently, and optionally drafts a reply. At 100 emails per day, that is 3,000 real-time inference calls per month. Fifty times the Weekend Reminder User's load.

Real-time inference is what costs Poke money. The company stated this without ambiguity: "What costs Poke money is real-time inference, like automations that run on every incoming email or real-time flight check-ins." At 3,000 calls per month, no free tier is viable. During beta, users with this profile reported negotiation outcomes in the $25–$30/month range. One widely circulated post on X described a starting offer of $292/month from Poke's bouncer AI, negotiated down to $29.

That $292 figure deserves a second look. It likely represents something close to the raw inference cost Poke incurs for a heavy-automation user — before the company applies the venture-subsidized pricing that growth-stage startups use to acquire users at a loss. The $29 negotiated outcome is the subsidized floor. The delta between $292 and $29 is being absorbed by Poke's $25 million in total funding.

For this persona, the relevant comparison is not Lindy at $21. It is the broader AI agent platform market, where per-conversation pricing runs from $0.50 (HubSpot, Quickchat AI) to $2.00 (Salesforce Agentforce) per interaction. At 3,000 inference calls per month, a per-conversation model would cost $1,500 to $6,000/month. Poke's subscription — even at the un-negotiated $292 — is structurally cheaper than per-call pricing at volume. That is the real value proposition for heavy users. It is also the reason the subsidy exists.

Monthly cost estimate: $19.99 to $30 on the current subsidized tier. The unsubsidized economic cost to Poke is almost certainly north of $200/month at this volume.

Scenario 3: The Integration Builder

Picture a third user — a developer who sees Poke not as a personal assistant but as a programmable interface layer. Poke supports webhooks, API access, and MCP (Model Context Protocol) for custom automations. This user does not text Poke "do I need a jacket today." This user wires Poke into a Slack workflow, triggers it from a CI/CD pipeline, or uses it as a natural-language front-end for a home automation stack.

The April 8 TechCrunch piece did not focus on this use case. But it is implicit in Poke's architecture — the company built Linq specifically to embed AI assistants within messaging platforms, and the webhook/API/MCP stack exists for exactly this kind of programmatic access. The infrastructure is there. The pricing is not.

The cost problem here is unpredictability. A developer building a Slack integration that routes every message in a 50-person workspace through Poke for classification would generate thousands of inference calls per day. A developer building a once-daily summary report would generate one. Same platform. Two orders of magnitude apart in compute.

During beta, this user profile was poorly served by the negotiation model. The bouncer AI set pricing based on stated usage intent, but developer workflows are inherently spiky — usage scales with adoption, not with the developer's initial estimate. A webhook that fires 10 times per day during testing fires 10,000 times per day in production. The pricing model has not publicly addressed this cliff.

Enterprise-tier AI agent platforms handle this with volume agreements. Salesforce Agentforce at $2/conversation is expensive for consumers but predictable for enterprise budgets. HubSpot's Customer Agent moved to $0.50 per resolved conversation in April 2026 — a concrete per-unit cost a finance team can model. Poke has not published enterprise pricing. The 6,000 VC insiders who were early users likely received bespoke terms. Those terms are not public.

Monthly cost estimate: indeterminate. The free tier works for prototyping. Production usage at scale requires a sales conversation that, as of May 2026, Poke has not publicly structured. This is where the $300 million valuation matters most — the company is pre-revenue-optimization, and pricing for power users is a gap in the public record that no amount of "as easy as texting" framing resolves.

Three usage patterns. Three cost profiles. One structural constant: the price you pay for Poke is a direct function of real-time inference volume, and the current pricing does not transparently surface that relationship.

The Weekend Reminder User may never pay a dollar. The Inbox Automator is getting a venture-subsidized deal that compresses a $200+ economic cost into a $29 monthly bill. The Integration Builder cannot price the product at all without a conversation that does not yet have a public on-ramp. All three users interact with the same text-message interface. The simplicity that earned Poke its headline — "as easy as sending a text" — is real at the interaction layer. It is not real at the billing layer.

This opacity is common across the 2026 AI agent market. Per-resolution pricing (Intercom Fin at $0.99, Zendesk at $1.50) makes the cost per interaction visible but punishing at volume. Subscription pricing (Poke, Lindy) makes the cost per interaction invisible but cheaper at scale. Neither model communicates the actual compute cost to the user. The difference is where the fog sits.

Poke's $25 million in total capital — $15 million seed from General Catalyst plus the $10 million extension — buys the company time to subsidize pricing while it grows. The 10x growth in user base that the team reported to TechCrunch suggests the subsidy is working as a growth lever. Whether it survives a Series A pricing correction is a separate question, and one that depends on how many of those 10x users are Weekend Reminder Users versus Inbox Automators.

Which Scenario Is You

The decision tree has three branches. Short ones.

If you text an AI assistant fewer than 10 times per day with minimal real-time needs, Poke's free tier or lowest paid tier is functionally equivalent to Lindy, ChatGPT, or any consumer assistant. Pick based on which messaging platform you prefer. The monthly delta is single digits.

If you process high-volume email or run continuous automations, Poke's current pricing is artificially compressed by venture subsidy. The value is real — $29/month for a service with $1,500+/month equivalent per-call market pricing — but that compression will unwind. Budget for it.

If you are building integrations via API, webhook, or MCP, Poke has not published pricing that covers your use case. Prototype on the free tier. Do not commit production workloads until a public rate card exists.

One number should anchor every decision here: $292. That was the un-negotiated starting price Poke's own system quoted to a user during beta — the closest public signal of what this service actually costs before venture capital absorbs the gap. Whether you pay $0, $19.99, or $29 today, $292 is the gravitational center the pricing moves toward once the subsidy window closes. Plan for it.

FAQ

How does Poke work without requiring an app install?

Poke uses Linq, a technology layer that embeds AI agents directly inside iMessage, SMS, and Telegram. You text a Poke contact, and the agent responds within your existing messaging app. No download, no account creation screen, no onboarding flow. The trade-off is platform dependency — WhatsApp support is limited after Meta barred third-party general-purpose chatbots in late 2025, narrowing the addressable messaging surface for non-Apple users.

What was the beta pricing negotiation and does it still exist?

During Poke's beta, new users encountered a "bouncer" AI persona that proposed a personalized monthly price based on their described usage. Users could negotiate downward. Reported outcomes ranged from $10 to $30 per month, with one documented case of a $292 starting offer negotiated to $29. This model has since been replaced with standardized pricing tiers starting at approximately $19.99/month for paid plans.

Who funded Poke and what is the current valuation?

General Catalyst led a $15 million seed round in 2025 at a $100 million valuation. An additional $10 million in 2026 brought the post-money valuation to $300 million — a 3x markup in under twelve months. Village Global and Earlybird participated, alongside angel investors from Stripe, Dropbox, and OpenAI. The company is registered as The Interaction Company of California, co-founded by Marvin von Hagen and Felix Schlegel.

How does Poke's cost compare to Lindy for light personal use?

Lindy charges $21/user/month for SMS-based personal assistant features, discounted to $18 through June 2026. Poke's paid tier starts near $19.99/month. At low usage volumes — a few texts per day, mostly reminders and calendar queries — the two are within $2/month of each other. The practical differentiator is messaging platform integration, not price. Both handle basic assistant tasks; neither is meaningfully cheaper than the other for casual users.

What types of requests burn through Poke's resources fastest?

Real-time inference is the primary cost driver. Requests that pull live data — weather, email triage, flight status, stock prices, smart home commands — require active compute and hit Poke's infrastructure harder than static queries like reminders, stored notes, or calendar lookups. A user sending five reminders per day consumes a fraction of the compute of a user routing 100 daily emails through automated classification. Poke's team confirmed that real-time automations are "what costs Poke money."

Does Poke offer API or enterprise pricing for developers?

As of May 2026, Poke has not published enterprise pricing or a public rate card for API, webhook, or MCP-based integrations. The technical infrastructure exists — webhooks, API endpoints, and MCP support are documented — but developers building production workloads cannot price the service from public information alone. The 6,000 VC insiders who accessed Poke early may have negotiated custom terms, but those arrangements are not available to general users.

Will Poke's current subscription pricing hold long-term?

Unlikely at current levels for heavy users. The spread between the $29 negotiated price and the $292 pre-negotiation starting offer during beta indicates substantial venture subsidy on per-user inference costs. With $25 million in total funding and rapid user growth, the company is optimizing for adoption over unit economics. A future fundraising round or revenue-optimization phase would likely compress that subsidy, pushing prices toward the un-negotiated range for users who generate high inference volume.