This review is part of a larger series of LinkedIn newsletters titled The Human Side of AI: Cutting through the AI noise to show you how AI can be a powerful tool for your creativity, efficiency, and strategy.
Scarlett is an agentic AI assistant that lives inside Slack. Built by Cracked AI and powered by Claude, it does more than answer questions. It runs tasks end-to-end from its own cloud machine, where it writes and runs code, pulls data from connected tools, and ships the finished artifact back into the channel. Cracked AI markets it as a coworker rather than a chatbot, and the framing mostly holds: Scarlett behaves like a junior operations teammate more than a research assistant.
For a market researcher, that distinction sets the whole review. Scarlett is strong at the plumbing around a study and thin at the analysis inside one.
Key strengths
Recurring desk research on a schedule
Scarlett can wake up on a set cadence and assemble a brief from sources the researcher chooses, mixing web search with connected tools. For competitive and market-landscape work, that means a standing Monday scan of named competitors, or a weekly pull of category news, delivered to a Slack channel without anyone re-running the query. Desk research that used to be a manual recurring chore becomes a scheduled task. The researcher sets the sources and the question. Scarlett keeps the cadence.
Data pulls and deliverable assembly
Much of a research deliverable is assembly: gathering numbers from several systems and shaping them into something a client can read. Scarlett connects to CRMs, billing tools, drives, and dashboards, then queries across them and builds the output as a shareable document or a dashboard. One operations reviewer noted it pulled quarterly figures from six tools in about a minute. For a researcher stitching survey exports and secondary sources into a findings pack, that assembly step is where the hours go.
Stakeholder and interview prep
Before a call, Scarlett can combine calendar detail, email history, and public web context into a preparation brief. For a researcher heading into a client readout or a buyer interview, that background gathering is real prep work, and Scarlett runs it automatically ahead of the meeting rather than the night before. Because it executes instead of only suggesting, the brief arrives finished.
Where it could improve
The gap that matters most for research is qualitative analysis. Scarlett is not built to code transcripts, tag themes across twenty interviews, or ground a finding in the exact quote that supports it. It executes and reports, but it does not synthesize evidence with traceable citations the way a qual study demands. For that work, a researcher is better served pairing it with a tool built for grounded analysis, such as NotebookLM or Claude used directly, and keeping Scarlett to the retrieval and assembly around them.
Data handling is the other open question, and for confidential work it is a gating one. The posture reads well on paper: isolated workspaces, encryption at rest, approval gates before high-risk actions, and a stated policy of not training on customer data. But at the time of writing, the compliance page lists SOC 2 as in progress, GDPR and CCPA documentation as in progress, and SSO as planned, with model requests routed through OpenRouter to outside providers. For transcripts under NDA, a lack of certifications is a real finding rather than a footnote. The product is also only weeks old, so the independent track record is thin. Pricing adds a smaller snag: the entry tier runs on a credit pool whose per-task value is not spelled out, and enterprise terms sit behind contact sales.
Ratings
| Dimension | Rating | Rationale |
|---|---|---|
| Usability | 4.0 / 5 | Slack-native setup in minutes and a talk-to-it-like-a-colleague interface put value within reach fast. |
| Power | 3.0 / 5 | Capable at data pulls and recurring briefs, but not built for the qualitative synthesis and grounded findings at the center of research work. |
| Flexibility | 3.5 / 5 | Broad tool integrations and scheduling, though the research-task range is narrow: no qual coding, survey design, or open-end analysis. |
| Cost | 3.5 / 5 | Fifty dollars a month is reasonable for an executing agent, but opaque credit economics and no standing free tier cloud the math. |
Conclusion
Scarlett earns a place in a researcher’s stack as an operations layer, not an analysis one. It handles the work that surrounds a study well: the recurring competitive brief and the deliverable assembled from scattered sources. The analysis stays human. The coding and synthesis stay with the researcher, along with the judgment about what the evidence means. Used that way, with sensitive transcripts held back until the security program matures, Scarlett can save real hours on the plumbing around a study.
Last updated: 8/5/2026.