Dust vs Genspark: Skills or Workflows?
Dust and Genspark both assign work to people and agents. Compare how they repeat steps, route approvals and charge for execution.
Founder, Task Machine
Should Dust or Genspark decide the steps of a recurring job through agent-selected skills or a configured Workflow? Both products let people and agents own tasks, so the choice depends less on whether work has an assignee and more on how a repeated request becomes an execution path.
Dust packages instructions and tools into skills that the agent chooses to activate. Genspark's Workflow builder creates a trigger and action steps that you can refine and test. For a weekly report, decide whether the agent should choose its method from context or whether your team needs to inspect the configured steps before switching the process on.
Dust starts with shared context
Dust centres collaboration on company knowledge, shared Pods and connected tools. Its Pod task documentation describes assignees and Open, In progress and Done states. People can complete tasks, agents can pick them up, and an agent's execution opens a linked conversation where a person can intervene.
That makes Dust a reasonable choice for a team whose work depends on shared context and ongoing conversation. A project-management Pod can contain an initiative's goal and success criteria, with scheduled summaries. The initiative example does not establish a named goal owner tracking a measurable target and acting when it changes.
For complex research, Dust also documents a research plan reviewed by a planning agent before specialist delegation. That is a specific planning process, not evidence that every Pod task goes through the same review.
Genspark puts creation beside coordination
Genspark combines agents with tools for documents, slides and spreadsheets. Its GenTeam guide describes tasks with one assignee, a status and a discussion thread. Agents claim work before starting, and people can claim the same kind of record.
GenTeam therefore overlaps with Dust on shared task ownership. A report-producing team should compare the creation tools and execution model, not assume one product is a task system and the other only a chat. Genspark also publishes a customer setup with a lead developer agent directing other agents, although that is a configured role, not a documented project-lead field.
The GenTeam guide describes both cloud agents and agents on a connected computer. Some local coding tools use the model access on that machine without Genspark credits. Hosted agents and local GenCode use Genspark credits, so the execution location alone does not tell you how the model is funded.
Who chooses the next step?
Dust's skills overview describes reusable instructions that can reference other skills. The agent decides which skills to activate. Its docs do not describe a native fixed-step workflow with defined branches that repeats the same execution path.
Genspark's Workflow guide starts with a repeated job described in chat, then creates a trigger and action steps. You can refine the configuration, test it and inspect each run's results and credit use. This gives a team a configured process to examine before routine execution.
For a weekly customer report, list the steps that must always happen, such as collecting the source material and preparing the output. If those steps can vary with context, Dust's skill selection may fit. If someone needs to inspect the configured sequence, start with Genspark's builder and test that sequence explicitly.
Genspark's test run changes real data and uses credits. Sends go out unless Approvals holds them, and configuration can itself execute real steps. Your evaluation should use a bounded example whose writes you intend to make.
Approval depends on the surface
Dust's changelog documents administrator-configured MCP tool approval levels, including always asking and never asking. An agent pauses for approval in its conversation. Dust also documents private Accept and Reject links for confirmed actions originating in emailed requests.
Genspark has separate mechanisms. GenTeam's Access tab controls use of connected apps, and requests outside those settings reach the creator as direct-message cards. The creator can approve once, grant broader permission or decline. In the Workflow builder, however, a send held by Approvals fails the run instead of opening a request for a decision.
For a report that includes an outbound email, check the exact surface that performs the send. A GenTeam access card does not establish that a builder run will pause for a human. Dust's tool approval settings also need to match the actual tool performing the action.
Runtime and payment are separate choices
Dust's Computer guide describes a temporary workspace for code and files. Its public docs do not establish an option to run the agent runtime itself on your computer or infrastructure. The custom-client tool preview is narrower than that runtime choice and should be evaluated as a preview.
Genspark's connected-computer option is documented in GenTeam. If your job needs tools or model access already present on a machine you control, that is a concrete reason to investigate it. Confirm whether you are selecting a supported local coding tool or GenCode, because their credit treatment differs.
Dust's pricing page lists Pro at €30 per seat monthly or €24 per month billed yearly, with 8,000 monthly credits, excluding VAT. Genspark's business pricing lists Team at $40 per seat monthly or $35 per seat per month billed annually, with 10,000 monthly credits per seat. Genspark reserves shared or distributed credit pooling for Enterprise. These figures describe different allowances, not a common unit of completed work.
Dust and Genspark, point by point
The table presents both products' answers to the same questions. Each answer includes its source and the date it was checked, so you can inspect the scope behind the label.
Both columns answer the same questions from each company's public pages. Yes means the company publishes the capability the question asks about, Partly means something narrower, No means it says it does not offer it, and Not published means its public pages do not describe it. Every answer links to its source and shows when we last checked it. The oldest check on this page was November 13, 2026.
| What you compare | Dust | Genspark |
|---|---|---|
| Goals |
Dust
Partly
A Pod can hold an initiative's goal and success criteria, with scheduled progress summaries. Its docs do not describe a named goal owner tracking a measurable target and acting on its changes. |
Genspark
Partly
AgentBase builds dashboards with key performance indicators, and GenTeam agents take named assignments. Its docs do not establish a goal owner tracking a measurable target and acting on progress toward it. |
| Leads |
Dust
Partly
Agents can coordinate project updates and delegate work to specialist agents. Its docs do not describe an assigned project, goal or team lead role responsible for reviewing and coordinating the work. |
Genspark
Yes
Genspark publishes a GenTeam customer setup where a lead developer agent directs other agents and hands implementation to a staging agent before human approval. This is a configured role rather than a documented project-lead field. |
| Tasks |
Dust
Yes
Pod tasks have assignees and Open, In progress and Done states. People can complete them, agents can pick them up and mark them done, and agent execution opens a linked conversation where you can intervene. |
Genspark
Yes
GenTeam tasks have a status, one assignee and their own discussion thread. Agents claim tasks before working, and people can use Claim for myself on the same task records. |
| Recurring work |
Dust
Yes
Schedule triggers run agents at times you describe, and webhook triggers respond to outside events. Each run creates a conversation, optionally in a shared Pod. |
Genspark
Yes
Workflows start on schedules, when Gmail or Outlook receives an email, or when meeting notes finish. GenTeam also schedules prompts for cloud agents and connected computers. |
| Workflows |
Dust
Partly
Skills package reusable instructions and can reference other skills, while agents delegate through Run agent. The agent decides which skills to activate. Its docs do not describe a native fixed-step workflow with defined branches that repeats the same execution path. |
Genspark
Yes
Describe a repeated job in chat and the Workflow builder creates a trigger and action steps. You can refine it, test it and turn it on, then inspect each run's results and credit use. |
| Autonomy |
Dust
Yes
Admins configure MCP tool approval levels, including always asking and never asking. Agents pause for approval in conversations, and actions requiring confirmation from an emailed request send private Accept and Reject links to the sender. |
Genspark
Yes
GenTeam's Access tab controls who may use each connected app. Requests outside those settings reach the creator as cards in a direct message, with one-time approval, broader permission or decline options. |
| Learned autonomy |
Dust
Partly
Medium-stake tools retain per-agent approval history, and self-improving skills propose instruction and tool changes from conversation feedback. Its docs do not describe proposals to change action permissions based on approval history. |
Genspark
Not published
Its public docs do not describe proposals to change an agent's permissions based on approval history. Access settings and a customer example of an agent-suggested release policy do not establish that learning loop. |
| Memory |
Dust
Yes
Agent Memory keeps user-specific information across conversations. You can inspect and delete saved memories in the agent's Memory tab. Each agent has separate memories for each user. |
Genspark
Yes
SecondBrain carries previous work into later agent requests and builds an About Me profile from projects, meetings and conversations. You can edit that profile directly and retain your edits through automatic updates. |
| Planning |
Dust
Yes
For complex questions, @deep-dive creates a research plan and consults a planning agent to review and improve it before delegating to specialist agents. You can follow the research reasoning as it runs. |
Genspark
Yes
The Workflow builder turns a conversation into configured steps you can test before switching on. Genspark Code also documents asking for a plan and waiting for approval, though that planning step is optional. |
| Intake |
Dust
Yes
Webhook sources receive events such as new tickets and route them to an agent's configured trigger. Workspace members can also forward email to [email protected]. Email from non-members is not processed. |
Genspark
Yes
A GenTeam cloud agent can receive instructions at its own email address from trusted contacts. Other senders' messages are held for the creator, and agent mailboxes are unavailable in the EU and Japan regions. |
| Connectors |
Dust
Yes
Dust offers built-in MCP tools, remote MCP servers and custom servers at a public URL. Built-in and custom webhook sources can start agents from events in connected services. |
Genspark
Yes
The connector catalog includes native APIs and Model Context Protocol servers, with organization policies for custom servers. Workflows can start from incoming Gmail or Outlook messages. |
| Models |
Dust
Partly
You choose a specific model or Basic, Standard or Premium. Auto uses a maintained ordered list with availability fallback. Dust explicitly says it does not select a different model based on each task. |
Genspark
Yes
Genspark describes automatic routing based on task type, context length, tool use and past performance. It also selects models for subagents, while GenTeam lets you choose Auto or a specific model per agent. |
| Cloud or your own computer |
Dust
Partly
Computer gives agents a temporary workspace for code and files. A preview supports tools executing in custom client applications. Its docs do not establish an option to run the agent runtime itself on your computer or infrastructure. |
Genspark
Yes
GenTeam offers hosted cloud agents and agents on your own connected computer. Genspark Claw likewise has a dedicated cloud option and a desktop mode that executes locally. |
| Your own subscription |
Dust
Yes
Dust's Gemini 4 Argon announcement explicitly includes workspaces that bring their own API key. That announcement does not specify eligible plans or the full set of supported providers. |
Genspark
Partly
GenTeam agents using Claude Code, Codex CLI, Cursor CLI or OpenCode on your computer use that machine's model access without Genspark credits. Hosted agents and local GenCode instead use Genspark credits. |
| Pricing |
Dust
Yes
Business offers Free seats with 500 lifetime credits, Pro at €30 per seat monthly or €24 per month billed yearly with 8,000 monthly credits, and Max at €150 monthly or €120 per month billed yearly with 40,000 monthly credits. Prices exclude VAT. Enterprise is custom. |
Genspark
Yes
Team costs $40 per seat monthly or $35 per seat per month billed annually, with 10,000 monthly credits per seat. Plus costs $24.99 monthly or $19.99 per month billed annually. Enterprise pricing is custom. |
| Correctness |
Dust
Yes
Dust documents PR reviewer agents that review pull requests against internal coding rules. This is a defined-criteria check for code reviews, rather than a published acceptance check on every Pod task. |
Genspark
Yes
Genspark describes grader agents for slides, sheets, documents, designs and coding projects. They assess artifacts and execution traces, including tool choice, detours and workflow delays, alongside user adoption and corrections. |
For software work
| What you compare | Dust | Genspark |
|---|---|---|
| Code review rules |
Dust
Yes
GitHub tools let you build PR reviewer agents that review pull requests based on your internal coding rules. |
Genspark
Partly
GenTeam documents asking a local agent to review a branch against main and return file and line references. Its docs do not establish code review against a configured set of repository rules. |
| Adversarial review |
Dust
Not published
Its public docs describe no independent second review of risky code changes. GitHub reviewer agents and sub-agent delegation do not establish that review process. |
Genspark
Partly
Genspark describes output-type grader agents, including for coding projects. Its docs do not establish an independent second code review triggered by risky changes. |
| Risk rating |
Dust
Not published
Its public docs describe no pre-merge risk rating for a code change. Published GitHub review tools and engineering use cases do not establish a risk score. |
Genspark
Not published
Its public docs do not describe a change-level risk rating before merge. The published model-routing risk tiers and database merge options address different jobs. |
| Pull requests |
Dust
Partly
GitHub tools can search, update, comment on and create pull requests, and agents can review them. Its docs do not establish agents revising a code branch in response to pull-request review feedback. |
Genspark
Partly
Genspark Code can edit an existing GitHub repository and push changes, and Claw documents reviewing pull requests. Its docs do not establish agents delivering pull requests and addressing their review feedback. |
A fact we cannot find or cannot check is left out, and prices can change after the date shown. Task Machine publishes this comparison and is an interested party. Dust and Genspark are named only to identify them, and there is no relationship between the companies. Has something changed? Tell us.
Who each one suits
Pick Dust if your team wants shared Pods, agent tasks and reusable skills guided by company context. Its honest limit for a prescribed recurring process is that the agent selects skills, and its docs do not establish a native fixed-step workflow with defined branches.
Pick Genspark if your team wants file-creation tools alongside assigned GenTeam work and configured Workflow steps. Its approval limit matters for outbound work: a held send in the builder fails the run, even though GenTeam separately offers access approvals.
Where Task Machine fits
Task Machine answers the execution-path question with Workflows that define ordered steps, branches, human questions and approvals. Verifiers can return failed results with feedback. Each Task also keeps its plan, decisions and history together, with acceptance requirements reviewed before execution. That fits a recurring job whose sequence and result checks must be explicit.
Task Machine is the wrong pick if your deciding requirement is Dust's shared Pod context or Genspark's integrated document, slide and spreadsheet creation tools, and those already meet the job. The Dust comparison and Genspark comparison separate those needs from reviewed execution. Start by mapping one outbound report in the Workflow documentation, including the approval that must happen before it is sent.