Carson vs Shortwave (2026)

The agent takes authenticated action in outside systems. It sends email to permitted recipients, posts slides to Slack, reads calendar events to trigger automations and pulls LinkedIn and research context into a unified workspace. All of it runs under a fine grained authorization framework enforced at the runtime.

SourceCarson launch post (vendor-authored, ycombinator.com/launches) 2026-02-26 and usecarson.com landing pageread 2026-09-08

The assistant writes into systems Shortwave does not own, through a named set of integrations. HubSpot lets it access customer information and update CRM data, and Notion lets it read, create and update pages and databases. Slack lets it access channels and send messages, Asana lets it create and assign tasks, Linear lets it track and update tasks, and Zapier connects it to thousands of other services. Each is turned on with one click from a settings page, and Arcade handles authentication.

Each write, whether updating a CRM record, creating and assigning an Asana task or sending a Slack message, is authenticated. Examples cross systems, such as "Create Asana tasks for each of the bug reports in my inbox" and "Find all emails related to Project Falcon and create a status update in a new page in Notion". Gmail, Google Calendar and Google Tasks are the native base, and the assistant creates calendar invites and edits events. Plans also add auto BCC for logging mail into HubSpot and Salesforce.

Tasklet, which connects to more than 3,000 apps, is a separate product at tasklet.ai. Shortwave integrates with it, but a Shortwave seat does not include it.

Sourceshortwave.com/docs/guides/ai-assistant AI Integrations and shortwave.com/pricingread 2026-09-13

One agent runs multi step knowledge work in phases. It does deep research into an editable spreadsheet, then builds a unified email workspace, then a branded deck edited in natural language. It can also create and run automations written in natural language, such as sending slides to Slack whenever a new intro meeting appears. Carson names no multi agent execution, named runtime or workflow builder.

SourceCarson launch post (vendor-authored, ycombinator.com/launches) 2026-02-26 and usecarson.com landing pageread 2026-09-08

The assistant takes one instruction and runs it as a chain of steps. In one example, "Find emails discussing the downtime last week, schedule a meeting with the people involved, and draft an FYI email about the meeting" runs as a three stage chain. It retrieves, then writes to the calendar, then composes, and each stage feeds the next.

Other examples chain retrieval into writes in other systems, such as "Search my email for recent product updates and create a blog post outline" and "Create Asana tasks for each of the bug reports in my inbox". The current assistant makes plans and takes multiple steps, running several AI powered searches and calendar lookups, backtracking to try different approaches and working around errors.

AI filters keep working after a request ends. They "automatically take action on new emails based on a custom AI prompt, including labeling, archiving, deleting, starring, and more". Plans include 3, 10 or 50 of them, and each is a standing rule that evaluates and then acts without a person present.

There is no workflow builder, no branching or conditional composition, no delegation between agents and no named runtime. The assistant assembles chains per request, and the lasting unit is a single condition filter. Tasklet's autonomous automation belongs to a separate product and is not part of this.

Sourceshortwave.com/docs/guides/ai-assistant, shortwave.com/pricing AI filters row and shortwave.com/blog/new-shortwave-ai-email-assistantread 2026-09-13

An automation written in natural language fires whenever a new intro meeting appears on the calendar and delivers slides to Slack. Carson names no schedule, webhook, inbound queue or wider set of events, and the in app prompt is the only other way to start work.

SourceCarson launch post (vendor-authored, ycombinator.com/launches) 2026-02-26 and usecarson.com landing pageread 2026-09-08

Mail arriving is the one event that starts the agent, through AI filters. They "automatically take action on new emails based on a custom AI prompt, including labeling, archiving, deleting, starring, and more", and plans include 3, 10 or 50 of them. Each filter runs unattended when new mail arrives, with the model in the loop. Auto apply labels and delivery schedules sit alongside, and instant AI summaries generate automatically on every thread.

The assistant is reachable on web, desktop, an installable PWA, iOS and Android, by Cmd-J or the sparkle icon on each, with granular push notifications. A person opens it in these places, and none of them starts work on its own.

Shortwave names no scheduled or cron runs, webhook, inbound API, external work queue or monitoring mode that starts an agent. Tasklet, which runs background automations around the clock, is a separate product at tasklet.ai that Shortwave integrates with, and its scheduler does not start Shortwave's agent.

Sourceshortwave.com/pricing AI filters and delivery schedules rows, shortwave.com/docs/guides/ai-assistantread 2026-09-13

Outputs are grounded on deep prospect research assembled for each task, such as a 10-K pulled for a deck, competitor pricing for a one pager, or LinkedIn and research for outreach. Carson names no persistent index, knowledge base or retrieval structure over the customer's own knowledge.

Sourceusecarson.com landing page and Carson launch post 2026-02-26read 2026-09-08

A maintained search index of the mailbox sits under the assistant. Pinecone, a vector database, is one of only four sub-processors, alongside Google Cloud, OpenAI and Anthropic.

How far back AI powered search reaches varies by plan, 5 years on Business and unlimited on Premier and Max, and it finds emails across the team's email history from a simple description instead of exact keywords. Each AI search draws on at most 50, 100 or 150 threads, depending on plan.

An about:[topic] query in the AI search infrastructure finds threads by description, not keyword, and an Import full history how to backfills the corpus. When the assistant answers from email it lists the relevant threads, with an about: button that opens the full set it drew on.

Sourceshortwave.com/docs/guides/security sub-processors, shortwave.com/pricing plans matrix and shortwave.com/docs/guides/ai-assistantread 2026-09-13

A Recent Work list holds a history of completed tasks across the organization. Carson names no memory or state the agent keeps across sessions, and it does not say how follow up context carries within a task.

Sourceusecarson.com landing page and Carson launch post 2026-02-26read 2026-09-08

AI Memories let a user steer the assistant by telling it what to remember. Memories "control the AI assistant's behavior, including how it responds, and what actions it takes along the way". Every paid plan includes the feature, offered as "Customize the behavior of your AI assistant by asking it to remember facts & preferences". Memories carry across sessions, and the assistant consults them as it works.

The user writes each entry in plain language. Every example is a standing instruction a person typed, from "Remember to use Oxford commas" to "Remember to add Asana tasks to the Marketing project". Shortwave names no way for the agent to write back what it learns during a run. It does not say where memories are kept or how a user would view, edit or delete one, and it publishes no retention, expiry or purge path for a memory apart from deleting the account.

Ghostwriter learns a writing voice from sent mail. Context carried from earlier searches into follow ups stays within one conversation.

SourceShortwave, shortwave.com/docs/guides/ai-assistant AI Memories section, shortwave.com/pricing and shortwave.com/changelogread 2026-10-05

The user tells the agent which email addresses it may contact and which require asking first. The runtime enforces that boundary, so the agent cannot act outside its grant. For recipients that need asking first, Carson prompts the user before acting, and the customer sets this per recipient and per action. Carson contrasts this with other agents that either ask approval for every action or work with no permissions at all.

SourceCarson launch post (vendor-authored, ycombinator.com/launches) 2026-02-26read 2026-09-08

Asking the assistant to organize the inbox produces recommendations instead of actions. "It will recommend actions you can take quickly on groups of threads like archiving low-priority emails, adding labels, deleting junk, and creating todos." Shortwave says this gets the user to inbox zero quickly "while still staying fully in control at each step".

An iOS release added the step where the assistant waits for confirmation while organizing the inbox. Drafts are always shown for review before sending and can be refined by follow up instruction. The assistant lists the threads it drew on behind an about: button, so the user can check what an answer was built from.

That confirmation covers inbox organization only. Through the integrations the assistant creates and assigns Asana tasks, sends Slack messages and updates HubSpot records, and Shortwave names no approval step for those writes. AI filters "automatically take action on new emails", including deleting and archiving, with no one confirming. There is no administrator layer, with no approval queue, policy engine, risk tiers by action or limits on what an assistant may touch. Oversight is per user and per surface, with no central governance.

Sourceshortwave.com/docs/guides/ai-assistant organize and integrations sections, shortwave.com/pricing AI filters row, and the vendor's App Store release notesread 2026-09-13

A fine grained authorization system enforced at the runtime lets a user grant the agent specific recipients and actions, and it blocks everything outside the grant. Carson publishes no attestation, trust page or security page, and it names no SSO or organization level role model.

SourceCarson launch post (vendor-authored, ycombinator.com/launches) 2026-02-26 and usecarson.comread 2026-09-08

CASA Tier 2 compliance covers the product, reviewed by third party security auditors and maintained by an annual security audit. For SOC 2 Type II and GDPR, Shortwave asks an organization that requires them to "contact our team" and claims no certificate.

Privacy and security in the plans come down to three features, CASA Tier 2 compliance, block spy trackers and privacy mode. Shortwave offers no SSO, SAML, SCIM, user provisioning, roles, groups, permissions model, admin console or customer visible audit log, and there is no enterprise tier for such controls.

Shortwave's audit logging covers its own employees. Access to customer data is tightly controlled internally and all of it is audit logged. No employee accesses data without explicit customer permission, and employee access requires hardware security keys. These internal controls are not something the customer's administrator operates.

Google Advanced Protection and the Google Workspace Marketplace listing are Google's programs, enforced by the customer's own Workspace administrator. Thread sharing, comments, assignees, shared labels and channels are collaboration features, not access controls.

Sourceshortwave.com/pricing plans matrix and shortwave.com/docs/guides/securityread 2026-09-13

A Recent Work list summarizes finished tasks by owner. Carson names no audit log, run trace or record of what the agent did and why. The runtime permission system enforces limits, and no log of that enforcement is shown.

Sourceusecarson.com landing page and Carson launch post 2026-02-26read 2026-09-08

Customers get no run history, activity log, admin dashboard, export, retention statement or audit trail of what the agent did. There is no customer facing record of assistant runs, filter executions or integration writes, including what AI filters deleted and archived unattended.

Shortwave's audit logging covers its own employees, whose access to customer data is very tightly controlled and audit logged, and only Shortwave reads that log.

Three features show the user something narrower. The assistant lists which threads it searched behind an about: button, which lets a user check one answer. Read statuses, recent opens and link tracking tell the user when a recipient opened an email or clicked a link. Shared labels are a team filing structure.

Sourceshortwave.com/docs/guides/security audit logging section and shortwave.com/pricing plans matrixread 2026-09-13

Carson is a desktop application installed on the customer's own machines, and the team offers to install it on site, so the agent runtime sits in the customer's environment. Carson does not say where model calls and data are processed or whether any server side component exists. No hosting region or hosting option is published.

SourceCarson launch post (vendor-authored, ycombinator.com/launches) 2026-02-26 and usecarson.comread 2026-09-08

All Shortwave data is stored in Google Cloud data centers, with AES256 at rest and TLS 1.2+ in transit. No region or region list is published, and customers cannot choose where data lives. Shortwave claims no residency of any kind, whether a country, a continent or an EU option.

Shortwave is delivered as cloud SaaS with native apps on iOS, Android, Mac and Windows. Where a customer installs the apps has no bearing on where the service runs. There is no on premises, private or dedicated tenancy, bring your own cloud or air gapped mode, and no enterprise tier that could carry one. The three plans are Business, Premier and Max.

The product is a Google Workspace Marketplace app that syncs a customer's Gmail into Shortwave's own Google Cloud estate and indexes it there, with OpenAI, Anthropic and Pinecone as named sub-processors.

Sourceshortwave.com/docs/guides/security data storage and sub-processors sections, shortwave.com/pricingread 2026-09-13

A rendering of the product home screen shows a Discover panel listing three named templates, Sales Discovery Workflow, Weekly Competitor Brief and Executive Research Template, beside the user's recent work. Carson does not say what selecting one does or whether the panel ships. Otherwise, interfaces and automations are generated on demand.

Sourceusecarson.com landing pageread 2026-09-08

Every reusable unit is one the customer writes, and nothing prebuilt ships. Users build saved prompts in the Assistant settings by choosing an emoji, naming the action and defining the prompt. The one example given, a coffee chat scheduler, is an illustration, not an entry in a library. Snippets let a user "create smart templates for writing common email content" and share them with a team, so the team library is whatever that team wrote. AI filters are natural language rules the customer writes, metered by count, and AI memories are instructions the customer writes.

Shortwave has no gallery, template library, starter pack, agent directory or marketplace. Its sharing tools carry only what users create. The named integrations are connectors, not agents.

Sourceshortwave.com/docs/guides/ai-assistant saved prompts section and shortwave.com/pricing snippets and saved prompts rowsread 2026-09-13

No model provider is named, and Carson names no choice of model for customers or admins.

Sourceusecarson.com landing page and Carson launch post 2026-02-26read 2026-09-08

The user chooses the model. Four intelligence levels each map to a named model. Basic runs Claude Haiku 4.5, Standard runs Claude Sonnet 4.6, Advanced runs Claude Sonnet 4.6 with adaptive thinking, and Expert runs Claude Opus 4.6 with adaptive thinking. The level is chosen per request, and "Quota usage varies based on the complexity of the request & the intelligence level selected."

Plans set the ceiling. Business tops out at Standard, Premier at Advanced and Max at Expert, and beneath that the choice is the customer's.

Every model on the selector is from Anthropic. OpenAI appears on the sub-processor list but is not a selectable model. Shortwave also says most of its AI workloads run open source models on hardware it controls, and those are not named or selectable.

Sourceshortwave.com/pricing AI Intelligence and quota rows, with shortwave.com/docs/guides/security sub-processors as corroborationread 2026-09-13

No API, SDK or MCP server for calling Carson from outside is published, and there is no documentation site or developer surface. The Slack and email connections let the agent reach out, and they are not a way in.

Sourceusecarson.com landing page and Carson launch post 2026-02-26read 2026-09-08

Customers can extend what Shortwave reaches with MCP servers, but no outside system can drive Shortwave. The assistant can "Connect to any tool with custom MCP servers", pointing to an MCP server of the customer's choice or one the customer writes.

Shortwave is the MCP client, not the server. The user provides the command, arguments and optional environment variables for a server and needs the latest desktop app, because Shortwave's own client starts the server as a local process. When a server fails, the logs are at ~/Library/Logs/Shortwave/main.log on Mac or %APPDATA%\Shortwave\logs\main.log on Windows. MCP support is limited to tool calls. Resources, prompts and sampling are unsupported, and only a limited number of tools (currently 40) can be active at once.

The documentation has four sections (Guides, Migrations, References and How-tos). Shortwave has no developer portal, REST or GraphQL reference, SDK, webhook, CLI or A2A card, and no docs or developer subdomain. The Google Workspace Marketplace listing is a distribution channel, not an API.

Sourceshortwave.com/docs/how-tos/using-mcp and shortwave.com/docs/guides/ai-assistant custom MCP sectionread 2026-09-13

Carson names no testing, evaluation or optimization tools for customers.

Sourceusecarson.com landing page and Carson launch post 2026-02-26read 2026-09-08

Customers get no test harness, scored cases, dry run, sandbox, regression check, A/B comparison of configurations or published quality metric.

AI filters are rules the customer writes in plain English that "automatically take action on new emails, including labeling, archiving, deleting, starring", and plans include 3, 10 or 50 of them. A new filter first runs against live mail, with deletion among its actions. Shortwave names no way to test that prompt against past threads, preview what it would have caught, or see what it did afterward.

A daily AI usage quota meters consumption. Intelligence levels let a user pay for a better model, and a follow up instruction refines an answer within one conversation.

Sourceshortwave.com/pricing plans matrix and shortwave.com/docs/guides/ai-assistantread 2026-09-13

As a desktop application, Carson generates its own interfaces, such as spreadsheets, email workspaces and slide editors, and acts through connected email, calendar and Slack. It names no browser, desktop or computer control, and it does not say how its deep research works.

Sourceusecarson.com landing page and Carson launch post 2026-02-26read 2026-09-08

The web browsing feature gathers information and does not operate websites. The browser is one of seven one click AI integrations, and its job is to "crawl the web to explore links and gather information". As AI web browsing in the plans, it lets the AI "browse the web to get the latest information related to your emails".

Shortwave names no hosted or local browser session that the agent drives, and no desktop control or computer control, remote or local. It reaches every system it touches through APIs. These are the Gmail and Google Calendar APIs as a verified Workspace Marketplace app, OAuth connectors to HubSpot, Notion, Slack, Asana and Linear brokered through Arcade, Zapier, and MCP tool calls for anything else. The desktop and mobile apps are Shortwave's own clients, and the agent does not operate anyone else's interface.

Sourceshortwave.com/docs/guides/ai-assistant AI Integrations and shortwave.com/pricing AI web browsing rowread 2026-09-13