Tab’s text-based AI assistant centers consent and payments in consumer agents

Tab, a new consumer AI assistant disclosed by TechCrunch on October 7, is betting that the next personal assistant will live not in a standalone app, but in a message thread. Users can ask it to buy groceries, book travel, call businesses, track tasks and manage everyday errands through iMessage or WhatsApp.¹

The product is notable less for its valuation than for its operating model. According to founder Ammar Amdani’s description to TechCrunch, Tab is designed with its own phone number, computer and wallet. That means the assistant is meant to take actions on a user’s behalf, rather than simply explain how to do them.¹ It also makes consent, payment custody and reversibility core product questions, not secondary privacy settings.

Tab’s public pitch places it in a fast-growing category of consumer agents that act as delegated operators. These tools do not merely answer questions. They enter websites, call businesses, make bookings, prepare transactions and return status updates in the same chat thread.²⁵ The interface is familiar — a text conversation — but the underlying permissions are closer to handing limited authority to a human assistant.

A message thread with operational tools

Tab works through iMessage and WhatsApp, letting users send requests in the same channels they already use for personal coordination.¹ Reports describe the assistant as able to order groceries, find gifts, make reservations, change bookings, call a hotel or doctor’s office, and handle other chores that usually require searches, forms or phone calls.¹⁵

The architecture behind that messaging interface is the central feature. CellCog’s analysis describes Tab as a text-based agent with three operational components: a phone number, a computer and a wallet.² The computer and browser environment would allow it to visit websites, use connected accounts and keep working while the user is not actively watching the conversation.²⁵ The phone number lets it place or receive calls for tasks that still depend on voice-based customer service.¹⁵

That design moves consumer AI away from the chatbot metaphor. KultureGeek framed Tab as part of a shift from assistants that provide advice to agents that perform actions for the user, including through a computing environment and payment tools.⁴ Blasting News Italia similarly described the product as an operational agent embedded in chat, with the user’s consent defining the boundaries of action.⁶

Approval before spending

The wallet is the most sensitive part of the product. Tab’s model appears to rely on an approval flow before money moves: the assistant presents what it is buying, the total cost and the payee, and the user must approve before payment is completed.²⁵

That approval step is central to the product’s trust claim. TechCrunch reported Amdani’s assertion that a user’s card is kept in a vault that Tab itself does not see, and that actions that cannot be undone require the user’s affirmative approval.¹ The Chinese-language outlet nstartup also highlighted the same trust positioning, including card vaulting, non-use of customer data for AI training and approval requirements before sensitive actions.⁷

For consumer agents, this is likely to become a standard design problem. A shopping assistant that can compare products is useful. One that can actually pay a merchant needs clear limits on amount, merchant, timing, refunds and records. Telset.id reported that Tab returns receipts to the message thread after an approved transaction and can continue tracking what happens next.⁵ That kind of audit trail may become as important as the initial approval button.

Data use becomes part of the product promise

Tab is also making a privacy claim at launch: User data is not used to train AI, according to TechCrunch’s report of Amdani’s comments.¹ For a personal assistant that may read messages, access documents, place calls, make purchases or interact with connected accounts, that claim addresses one of the category’s most immediate risks: whether intimate life-administration data becomes training material.

The claim remains a product promise that readers should distinguish from an independently audited technical disclosure. CellCog noted that, as of October 7, Tab had not published pricing, the underlying model, or extensive public documentation such as a detailed security page or privacy summary that could be compared with the founder’s statements.² Agent Sonic’s analysis similarly emphasized gaps in Tab’s public disclosures, including pricing and implementation details, while noting the broader trust questions raised by wallet-based delegated errands.³

That gap matters because agents that operate across websites, messages, payments and calls create a wider consent surface than ordinary chatbots. Users need to know what the assistant can access, what it stores, which third-party services process requests, how long records are retained and how to revoke permissions.

Reversibility is becoming a consumer feature

The most important product line in this category may be between reversible and irreversible actions. A reminder can be deleted. A draft message can be rewritten. But a purchase, cancellation, booking change or call to a medical office can create real-world consequences.

Tab’s stated approach — requiring approval before actions that cannot be undone — reflects a broader design principle for consumer agents.¹⁷ The more capable the assistant becomes, the more the product must show decision points to the user: when it is gathering information, when it is preparing an action, when it is about to commit money and when it has already acted.

Competitors and adjacent tools are exploring similar boundaries. Agent Sonic, which operates in WhatsApp, says its outreach tasks show the exact first message before sending and require a user reply before contacting someone.³ While not identical to Tab’s broader wallet-and-computer model, that pattern points to the same consumer expectation: Delegated agents should show their work before they act.

The race shifts from demos to trust mechanics

Tab is entering a crowded market that includes other consumer assistants such as Instinct and Meta’s Muse, as well as messaging-based tools focused on reminders, follow-ups and admin work.¹³ But the competitive question is increasingly architectural. The most successful products may not be the ones that produce the most impressive demos, but the ones that make delegated action understandable, bounded and reversible.

For users, the key evaluation criteria are practical: Can the assistant complete tasks reliably? Does it ask before spending money or contacting people? Are payment credentials separated from the agent’s own systems? Is there a record of actions and receipts? Can permissions be revoked? And are data-use promises written into policies rather than left as launch-week statements?

Tab’s launch shows where consumer AI is headed. The personal assistant is becoming less like a search box and more like a limited operator with tools, accounts and authority. That could make everyday administration easier. It also raises the threshold for transparency, because a tool that can act for a user must make clear when, how and why it is acting.