What is Writer?
Writer—styled WRITER by the company—is an enterprise AI platform for agentic work. It has moved well beyond its origins as a writing and brand assistant. The current product centers on Writer Agent, which can plan multistep work and generate documents, presentations, spreadsheets, dashboards, analysis, or actions in connected systems. Users can save repeatable instructions as Playbooks, schedule Routines, ground results in a company Knowledge Graph, and build more specialized agents in AI Studio.
Writing remains an important capability. Brand personalities, terminology, style guides, approved knowledge, and departmental rules can shape output. But a current evaluation should compare Writer with enterprise automation and agent platforms, not only grammar or copywriting tools. Browser automation, code execution, web search, data analysis, connectors, APIs, and cross-team workflows widen both its value and its operational risk.
Agent, Playbooks, and company knowledge
Writer Agent provides the general interface. A one-off instruction can become a shareable Playbook with defined steps and expected output, and Routines can run work on a schedule. Knowledge Graph connects documents and enterprise sources so an agent can retrieve company context. Enterprise options add advanced connectors, orchestration, approvals, model flexibility, access controls, observability, and auditability.
Grounding does not guarantee truth. An agent can retrieve an outdated policy, miss an exception, combine sources from different jurisdictions, misunderstand a table, or cite a passage that does not support its conclusion. Brand rules can make an answer sound approved even when its facts are wrong. Establish an owner, approved source set, effective date, exclusion rules, and acceptance criteria for each Playbook.
Start with read-only, reversible work: summarize a controlled source set, draft a launch brief, compare contracts into a review table, prepare a meeting pack, or classify support cases without sending a response. Capture the baseline time, correction rate, citation accuracy, accepted output, and downstream outcome. Only add actions after the read-only workflow meets its thresholds.
Plans, credits, and total cost
Writer's current public plans page offers Starter and Enterprise. Starter has a 14-day trial without a credit card, supports up to five users, five Playbooks, three scheduled Routines, a single limited Knowledge Graph with 50 GB storage, up to three active basic connectors, a team personality, and limited AI Studio access. Enterprise advertises unrestricted workflows, broader connectors, governance, support, and as many users, teams, and free collaborators as needed.
The page does not currently show a numeric Starter seat price, Enterprise price, or included credit allocation. It says Starter uses monthly or annual per-seat plans with fixed credit limits. Enterprise combines seats, free Lite users, optional solution packs or services, and an organization-level credit pool. Unused enterprise credits expire at the end of the contract term, and current help documentation says spending-limit features are still developing.
Request a cost model for Pro and Lite seats, initial credits, every model and tool, code, browser, files, search, Knowledge Graph storage and refresh, API, embedded users, connectors, professional services, support, overages, and renewal. Measure cost per approved completed workflow, not per generated word. Include knowledge curation, identity, integration, evaluation, monitoring, incident response, change management, and exit.
Customer data, retention, and third-party models
Writer's enterprise Platform Services Agreement says the customer retains Customer Materials and owns Output to the extent permitted by law; Writer assigns any output rights it has to the customer. It licenses customer material only to host, store, back up, transmit, display, and process it to provide and support the platform, and says it will not use Customer Materials to train its models. Output can still resemble other output, and the customer remains responsible for accuracy, appropriateness, and use.
The DPA describes Writer as processor, restricts personal-data use to the business relationship and agreement, prohibits selling, sharing, or targeted advertising with covered personal data, provides subprocessor notice and objection mechanics, and permits return or deletion on request subject to law. The individual privacy policy governs website visitors and personal-email accounts, while an organization's agreement controls enterprise processing.
Writer describes a “zero data retention approach,” meaning it stores data only as long as needed to build with the platform. Do not read that as “nothing is stored.” A 50 GB Knowledge Graph, uploaded files, chat sessions, generated documents, agents, workflow state, audit logs, backups, and connector caches necessarily have lifecycle rules. Writer also supports preferred third-party LLMs on Enterprise. Confirm which model processes each workflow, whether data leaves Writer, provider retention, region, subprocessors, feedback, telemetry, backups, legal holds, and deletion verification.
Agent security and governance
An agent connected to email, CRM, cloud storage, browsers, code, or internal tools can leak data or take the wrong action. Retrieved documents and websites may contain prompt injection telling the agent to ignore rules or exfiltrate secrets. A scheduled error can repeat at scale, and a compromised connector can turn a content workflow into an access path.
Separate build, test, and production. Use dedicated service identities, least privilege, allow-listed destinations, short-lived credentials, per-tool scopes, secret isolation, rate and cost caps, and human approval before sending, publishing, changing a record, executing code, making a purchase, or contacting a customer. Test indirect prompt injection, malicious attachments, data poisoning, cross-team access, broken permissions, ambiguous instructions, and model changes.
Log the prompt or Playbook version, sources, connector calls, code, inputs and outputs, approvals, model, cost, and final action. Give every workflow a business owner, technical owner, data owner, rollback, kill switch, incident path, review cadence, and retirement date. Audit logs are valuable only if someone reviews and can act on them.
Verdict
Writer is a serious enterprise platform for organizations that want company-grounded agents and repeatable work under a central governance layer. It is particularly relevant when brand, knowledge, workflow, and compliance must be managed together rather than scattered across consumer AI accounts.
Its public pricing is not transparent enough for a purchase decision, and its agents should not receive broad production authority from day one. A strong rollout uses a representative read-only pilot, written unit economics, contract-level data review, adversarial testing, least privilege, and measured expansion based on approved outcomes—not the volume of generated content.
