Obviously AI and Zams AI Workers

Obviously AI is now Zams, and the product has changed with the name

Anyone arriving at Obviously AI with memories of no-code predictive analytics needs to pause before comparing old features or prices. The company now says plainly that Obviously AI is Zams, while the obviously.ai site is an archive of how the earlier product operated through 2025. The active business sits at zams.com and sells AI workers and a multi-agent automation platform for business teams. This is not a cosmetic rebrand around the same dashboard. It is a change in the job the product is asking to do.

Obviously AI originally helped non-technical users train predictive models from tabular data. Classification, regression and time-series forecasting were the center of the experience. Zams moves from prediction toward execution. It connects business systems, accepts instructions in ordinary language, searches and reasons across company context, and uses agents to complete tasks. The most visible use cases now sit around revenue operations: meeting preparation, CRM hygiene, account monitoring, reporting, renewal risk and call coaching.

That transition is the essential fact for a buyer in 2026. Old Obviously AI reviews, plan names and model-building tutorials can still explain the company's history, but they do not reliably describe the product being sold now. The active evaluation should be Zams versus enterprise automation and agent orchestration platforms, not Obviously AI versus traditional AutoML tools.

The product in one sentence

Zams is an enterprise automation platform from Obviously AI that coordinates AI agents across connected business apps, with ready-made AI workers and recipes for research, reporting, data maintenance and operational tasks.

The phrase AI worker is Zams' preferred framing. The company distinguishes it from a chatbot or copilot that returns text for a human to finish. In the Zams model, a worker receives a goal, gathers context, plans steps, acts across tools and reports back. That is an ambitious promise, and it should be tested at the level of completed actions rather than polished answers.

From predictive models to operational action

The old Obviously AI product reduced the friction of building machine-learning models. A user connected a dataset, chose an outcome, trained a model and explored predictions. Zams keeps some of that data-oriented DNA, especially in forecasting and reporting, but shifts the center of gravity to work that crosses several systems.

Consider a common revenue operation. Preparing for a customer meeting can require calendar details, CRM history, recent emails, public company research, call transcripts and current account activity. A dashboard can display parts of that information, and a copilot can summarize whatever is pasted into a prompt. Zams is designed to connect the sources, retrieve the relevant context, assemble the briefing and deliver it as a repeatable operation.

The same pattern applies to CRM hygiene. A conventional automation might say that when one field changes, another field should be updated. An agentic system is supposed to reason about incomplete records, find missing context, choose an action and update the system. Zams describes a multi-agent architecture and a Z1 Engine that combines orchestration, persistent memory and contextual reasoning for this kind of work.

The important word is supposed. These are vendor descriptions, not an independent guarantee that every messy enterprise workflow will run correctly. Agentic automation can fail in more consequential ways than a bad summary because it can change records, send messages or trigger downstream activity. A responsible evaluation measures completion, accuracy, exception handling and recoverability with the buyer's own systems.

Meet the current AI workers

Zams now presents named workers for specific jobs. Evan is the clearest commercially available example on the public pricing page. It researches the people in an upcoming meeting and prepares a briefing. Evan is listed at $50 per month through a separate product experience.

The wider lineup shows where Zams wants to take the category. Iris is positioned around relationship intelligence and warm introduction paths. Nico monitors accounts for buying signals. Nova profiles successful customers and finds similar prospects. Atlas focuses on CRM search, updates and data quality. The public site currently labels Iris, Nico, Nova and Atlas as coming soon, while the company's June 2026 article says Evan and Atlas are the start. Buyers should therefore separate the roadmap from the capabilities they can deploy today.

This matters because a list of named workers can create the impression of a finished digital department. The current buying question is narrower: which worker or recipe is live, which integrations are operational for that use case, what permissions does it need, and how does the team review its work? Ask for a demonstration using a realistic account, not a generic sample.

The automation platform underneath

Beyond individual workers, Zams sells an automation platform. Its pricing is based partly on agent tasks, enrichment tasks and PDF pages, with plan differences around users, support, governance and deployment controls. The public recipes show the practical orientation: CRM hygiene, quota forecasting, renewal-risk alerts, executive briefings, deal intervention, performance reports, automated metrics, call coaching and lead intelligence.

The platform pitch is natural-language configuration instead of brittle if-then builders. A team connects its stack, describes the intended outcome and lets agents coordinate the steps. That can be valuable when a process cannot be expressed as one fixed trigger and one fixed action. Research, interpretation and conditional decisions are exactly where agentic automation may add more than a traditional connector.

It also creates a design challenge. Plain language is easy to write but can be ambiguous. A request such as keep our CRM clean hides dozens of decisions about authoritative sources, matching rules, field ownership, acceptable confidence and what to do when systems disagree. The best deployment converts that broad sentence into an explicit operating policy. Natural language should make configuration easier, not make accountability vague.

Integrations are the real product boundary

Zams says it can work across email, calendars, CRMs, drives and databases, and its pricing FAQ names Salesforce, Slack, Notion, Google Workspace, custom tools, ERPs, internal APIs and databases. The main site also describes reporting from systems such as Salesforce, HubSpot and Gong.

However, the public integrations directory currently marks many named connectors as coming soon. That is a reason to verify, not a reason to dismiss the platform. Public directories can lag private deployments, and enterprise integrations may be enabled through services or custom connections. Still, no buyer should assume that every logo represents a self-serve, production-ready connector.

Build the evaluation around your exact stack. Ask whether the connector is generally available, beta, custom or roadmap. Confirm read and write operations separately. Check how authentication works, whether service accounts are supported, how fields are mapped, how rate limits are handled and whether actions are idempotent. A platform that can read Salesforce but cannot safely update the records you need is a different product from one that can execute the complete workflow.

Pricing has become enterprise-shaped

The old Obviously AI prices circulating in comparison articles are historical and should not be used for Zams. The active Zams pricing page currently lists Evan at $50 per month. For the automation platform, Basic is $125 per month, Pro is $1,208 per month and Team is $1,417 per month at the visible default configuration. Enterprise is contact-based.

Basic includes one user, 1,000 agent tasks, 10 enrichment tasks and three PDF pages in the displayed configuration. Pro shows one user, 50,000 agent tasks, 1,000 enrichment tasks and 100 PDF pages. Team shows up to 25 users with the same displayed task and page allowances. Support and governance capabilities vary by plan, including email or live-chat support, dedicated channels, training, priority support, role-based access, observability, single sign-on and data residency.

The page includes a volume selector and monthly or annual options, so the visible figures are not a universal quote. Zams says annual billing receives a discount, and enterprise requirements are customized. A guided trial can be requested, but the pricing page does not present an unrestricted permanent free plan.

Agent tasks deserve particular attention. Zams defines one as an action performed by an agent or one API call used to interact with an agent. That means a business outcome may consume multiple tasks. Updating a CRM, sending a message and writing a report can each be separate billable work. Before comparing plans, map one real workflow step by step and ask Zams to estimate the task, enrichment and document usage under normal and retry conditions.

Who should put Zams on the shortlist

Zams is best suited to revenue operations, sales enablement, customer success, finance, data and operations teams with a meaningful software stack and a repeated process that crosses several systems. It becomes more interesting when people spend time finding context, updating records, assembling reports or preparing for interactions, rather than doing work that requires a single isolated app.

The buyer should have an accountable process owner. Agent platforms are not a shortcut around process design. Someone must define what correct looks like, identify permitted actions, own exceptions and decide when a human must approve. The strongest pilot has enough repetition to measure, but not enough risk to cause serious damage if the agent behaves unexpectedly.

Zams is less suitable for an individual who simply wants personal task automation at consumer pricing. Basic starts at a professional platform price, and the more capable plans are enterprise investments. It is also no longer the obvious choice for someone whose sole need is no-code predictive modeling from a spreadsheet. The old Obviously AI material may describe that job, but Zams should now be judged as automation infrastructure.

The implementation work buyers should expect

Start with one workflow, not an AI workforce. Meeting preparation is a sensible pilot because the output can be reviewed before it affects a system of record. Define the triggering event, the approved data sources, the fields that matter, the delivery channel and the deadline. Create a small evaluation set of meetings with known context and compare the generated briefings with what a strong human researcher would prepare.

Measure missing facts, incorrect facts, irrelevant detail, source traceability and delivery reliability. Ask whether the worker distinguishes verified data from inference. If public research is involved, establish which sources are allowed and how personal information should be handled. The goal is not to be impressed by one rich briefing. The goal is to know whether the worker performs consistently across ordinary and difficult cases.

For a write-enabled pilot such as CRM hygiene, add more controls. Begin with read-only analysis or suggested changes. Require approval for bulk edits, deletions, ownership changes, customer communication and financial fields. Log the previous value, new value, evidence and responsible agent. Test duplicates, missing identifiers, conflicting systems, rate limits and partial failure. Make rollback part of the design before the first production update.

Cost measurement should run alongside quality measurement. Count agent tasks, enrichment calls, document pages and human review time. Include the cost of integration, policy design and exception handling. Vendor ROI calculators can suggest a hypothesis, but only the pilot can show whether the organization actually reclaims time or simply moves work into supervision.

Security and data governance are serious strengths, with caveats

Zams publishes a substantially more enterprise-oriented security posture than many lightweight automation tools. The company states that it is SOC 2 Type II, supports GDPR and CCPA obligations, is prepared to execute a HIPAA business associate agreement, and uses Google Cloud infrastructure. It describes encryption in transit and at rest, access logging, multi-factor controls and data deletion rules.

Its April 2026 privacy policy says customer data is processed to provide the service and is not used to train proprietary or third-party AI models unless the customer authorizes that in writing. It also says agent outputs and prompts are not repurposed for general training or cross-customer insights, while aggregated and anonymized metadata may be used for platform improvement. The terms say customers retain rights to their data, prompts, configurations and outputs.

Those are meaningful statements, but enterprise diligence still belongs in the contract. Ask for the current trust documentation, data processing agreement, subprocessor list, model-provider details, retention schedule, deletion procedure and incident commitments. Confirm whether connected data is stored, cached or sent only during active sessions for the specific deployment. Determine what the published statement about recordings means for your chosen worker.

The terms also say users should not process specially protected data, including health information or financial account data, without prior written consent. A healthcare or finance buyer should not treat a compliance badge as blanket authorization. Obtain the appropriate agreement and written scope before connecting regulated systems.

Permission design is the heart of safe agents

An agent should receive the least access required for its job. A meeting researcher may need calendar and CRM read access but no ability to send email. A CRM maintenance agent may need write access to selected objects and fields, but no permission to delete accounts or export the full database. Separate service accounts make actions easier to trace and revoke.

Approval workflows should be based on consequence, not novelty. Low-risk, reversible updates can become autonomous after evidence accumulates. High-risk actions should retain human approval even when the agent has performed well. These include external messages, pricing changes, financial commitments, legal records, access changes and mass edits.

Observability should answer four questions: what goal was given, what context was used, what actions were attempted and what actually changed. If a buyer cannot reconstruct an incident, the automation is not ready for important operations. Ask to see logs during the sales process, including a failed or interrupted run, rather than viewing only successful output.

How Zams compares with familiar alternatives

Zapier is the obvious reference point because it has a large connector ecosystem and a mature trigger-action model. Zams is trying to move beyond fixed workflows into goal-driven, multi-step execution with business context. Zapier may be preferable when the process is deterministic and connector availability matters most. Zams becomes interesting when research, reasoning and orchestration are central.

Relevance AI and Lindy occupy nearby agent-building territory. The practical difference will not be the word agent on a landing page. Compare the supported systems, configuration model, governance, pricing meter, ready-made use cases and implementation support. A platform that fits the buyer's identity model and approval rules can be more valuable than one with a longer capability list.

Traditional business intelligence tools remain relevant too. Zams can answer questions and generate reports from connected data, but a governed BI layer may still be the right source for certified metrics. An AI worker can distribute or explain a measure without becoming the authority that defines it. Decide where metric logic lives before automating executive reporting.

The WhatAI verdict

Obviously AI's move to Zams is one of those transitions that makes old reviews actively misleading. The company has moved from helping users build predictive models to selling AI workers and an agent automation platform. The correct WhatAI category is now Automation, with a strong sales and revenue-operations orientation, not Analytics as the primary label.

The product has a clear idea: business software should complete work across the stack, not merely display information or draft an answer. Evan gives that idea a focused entry point, while the automation platform and recipe library address broader operational jobs. The public roadmap is ambitious, but several named workers and integrations are still marked coming soon, so capability verification must be specific.

Put Zams on the shortlist when your organization has a costly cross-system process, enterprise governance requirements and a team willing to design a measured pilot. Do not buy an abstract promise of digital labor. Buy one defined outcome with known inputs, permitted actions, acceptance criteria, logs and a rollback path. If Zams can deliver that repeatedly, the larger AI-worker vision becomes credible. If it cannot, a simpler workflow tool may be the better answer.