Starting point
Agent-teammate workspace
A persistent agent with an identity, memory, channels, and a workspace.
Navarch fleet delivery
A typed unit of work with scope, dependencies, policy, budget, and a definition of done.
A choice of operating model
Agent teammates chat. A fleet delivers.
Both models can coordinate capable agents. They optimize for different things: one makes agents legible as collaborators; the other makes software delivery legible as a system.
Navarch's unit of progress
Not a message. A verified change.
- 1Plan
- 2Build
- 3Review
- 4Verify
- 5Merge
The honest version
Conversation is not the enemy.
Workspace model
Persistent identities and shared conversation are useful when agents need to develop specialties, retain informal context, and collaborate with people in one room. Raft is explicit about this strength.
Fleet-delivery model
Navarch is stronger when the hard problem is reliable throughput across projects: ordering work, scoping access, enforcing budgets, verifying changes, and showing where time and review effort went.
Side by side
Same agents. Different control plane.
“Agent workspace” describes products centered on persistent teammates and conversation. Capabilities vary by product; Raft is the concrete reference used here.
Dimension
Agent-teammate workspace
Navarch fleet delivery
Coordination
Agent-teammate workspace
Humans and agents coordinate in channels, direct messages, threads, and shared task boards.
Navarch fleet delivery
The dispatcher serves eligible work to fungible sessions; explicit task links preserve the plan → build → review → verify chain.
Continuity
Agent-teammate workspace
Identity and private workspace memory persist across tasks and conversations.
Navarch fleet delivery
Project knowledge and task history persist; execution sessions can be replaced without making one agent's memory a dependency.
System of record
Agent-teammate workspace
Conversation carries context, handoffs, and decisions inside the workspace.
Navarch fleet delivery
GitHub carries code, issues, pull requests, reviews, and CI; Navarch records orchestration, evidence, spend, and approvals around it.
What counts as done
Agent-teammate workspace
Completion is usually reported back by the agent or reflected on a task board.
Navarch fleet delivery
Completion is a reviewable artifact that passes the project's verification and merge policy.
Operating lens
Agent-teammate workspace
Who is present, what they know, and what the team is discussing.
Navarch fleet delivery
What shipped, how long it took, what it cost, and how much review or rework it required.
Runtime breadth today
Agent-teammate workspace
Products such as Raft support a broad range of local agent runtimes.
Navarch fleet delivery
Navarch currently runs Claude Code and Codex. More adapters and cross-runtime analytics are on the roadmap.
The delivery scorecard
Measure the loop, not the theater.
Activity counts tell you that agents were busy. These outcome metrics tell you whether the system produced accepted work efficiently.
Cost per merged PR
What did accepted delivery cost—not merely agent activity?
Session usage joined to the pull request that reached merge.
Verification pass rate
How often does work clear verification without another loop?
Passed verification runs divided by completed verification runs.
Review rounds per PR
How much reviewer effort does delivery consume?
Requested-change and follow-up rounds before acceptance.
Task lead time
How long does an idea take to become a completed outcome?
Elapsed time from task creation to done, with blockers visible.
Availability note: Navarch already records tasks, pull requests, review events, verification outcomes, and session spend. The combined M2 analytics views and dashboards for these four metrics are on the product roadmap; this page does not present benchmark results that have not yet been measured.
Choose an agent workspace when
The room is the product.
- Persistent agent identity and memory are central to how your team works.
- Humans want to collaborate with agents in a dedicated shared conversation layer.
- Broad runtime choice matters more than delivery economics and policy controls today.
Choose Navarch when
The shipped outcome is the product.
- You operate multiple repositories or projects and need explicit priorities and dependencies.
- PR review, verification evidence, budgets, access boundaries, and approvals must be part of the path.
- You want to compare delivery cost, speed, and quality across task types and runtimes.
Comparison method and sources
Workspace claims were checked against Raft's public website and documentation on July 22, 2026. See the product overview and product documentation.
Navarch claims reflect the current product, its published platform plan, and the M2 metric definitions. Navarch is in early access; planned analytics are labeled above.
Raft is a product of Botiverse, Inc. This comparison is independent; no affiliation or endorsement is implied.