Ando's Slack Pitch Starts With a $3,480 Baseline • The Weighted Average

Small engineering teams should test Ando on a real handoff before replacing Slack after the startup launched its human-and-agent messaging platform with $20M in funding on September 24. At the upper end of its initial target cohort, the incumbent Slack Pro subscription benchmark is $3,480 a year—large enough to price explicitly, but too small a denominator to justify migration without evidence that coordination improves.

The subscription is the smallest moving part

Ando’s launch essay targets teams of 2–40 humans and says pricing is per human seat. It does not publish a dollar tariff. The company says it does not want customers counting agent messages, actions, or conversations, while leaving room for pricing to evolve. That is a proposed charging philosophy, not a guarantee that outside model services, an existing harness, or future commercial terms have no cost.

The incumbent benchmark is reproducible. Slack lists Pro at $7.25 per user per month when billed annually. Multiply that rate by Ando’s 40-human initial ceiling and twelve months: 40 × $7.25 × 12 = $3,480. This combines the new vendor’s target team size with the incumbent’s public price. It is a normalized US list-price comparison, not a claim about an actual Ando customer, a migration quote, or money already saved.

The plan matters as much as the headcount. On the same pricing page, Business+ costs $15 per user per month annually, producing $7,200 for the same cohort. That is $3,720 above Pro. The difference cannot be called an Ando saving: the plans include different capabilities, and Ando’s public launch material does not establish feature equivalence. It does show why asking merely whether a new messenger is cheaper skips the question of what must be replaced.

Nor does a switch automatically release the full subscription benchmark. Slack’s Fair Billing Policy distinguishes active paid members from free bots and single-channel guests. Under that policy, eligible inactive-seat credits are nonrefundable and apply toward future payments. A team keeping Slack for customers, contractors, or old conversations may continue paying. Use the actual contract and seat activity, not the entire modeled annual bill, as the avoidable-cost line.

Ando’s opportunity is therefore not simply to remove a software charge. It is to remove the human who repeatedly carries context between an agent and the rest of the team. TechCrunch describes that relay problem and Ando’s ambition to replace existing internal messaging. If fewer handoffs shorten accepted work, the subscription may be secondary. If the team merely acquires another inbox, even a cheap seat becomes expensive coordination.

The best first customer is a small team already using agents on shared work, not one hoping a new chat application will create that practice. Ando says it starts with smaller teams and plans to onboard larger groups later. Treat that cohort choice as part of the product’s current fit. A broad enterprise migration needs evidence about controls, continuity, and support that a successful small-team demonstration cannot supply by itself.

Shared context needs an owner

The product changes who participates in a conversation. Ando’s announcement describes agents with persistent identities, permissions, shared context, and access to channels, threads, and live conversations. Its launch essay says agents can join conversations and pick up work without being explicitly tagged. That is a different interaction model from asking a private assistant for an answer and manually forwarding the useful portion to colleagues.

The examples make the proposition concrete. Ando describes agents handling bugs, supplying context to product discussions, preparing changes, and reviewing one another’s work. It also says teams can bring different agent harnesses and offers a hosted native option for users without a preferred harness. These are vendor descriptions of available workflows, not independent measurements of productivity. The purchasing test should preserve that distinction rather than promote an attractive demonstration into a throughput claim.

Accel’s investment account argues that coordination becomes harder as context, memory, and activity spread across more agents. As an investor, Accel has an evident interest in that thesis. Still, it identifies a plausible problem: individually capable agents can leave people managing the seams. A common workspace may make those seams visible, but visibility alone does not settle ownership when two participants act on different understandings of the same request.

Test the seam rather than the greeting. Choose work in which an agent must learn a changed requirement, ask another participant for missing information, and return an artifact that a named human accepts. Observe whether the final result follows the latest approved decision. Count the clarification and correction work as part of the task. These are proposed pilot criteria, not a claim that Ando has failed them or already passed them.

There is an important distinction between being present and being authorized. An agent may need to read a discussion without having permission to change a production system. A channel invitation should not silently become a grant over every tool attached to its harness. Keep the action boundary explicit in the underlying application or execution environment, and make its owner responsible for proving what can happen without another approval.

Our AI Employees analysis separated role descriptions from the permission boundary enforced by the runtime. The lesson transfers without equating the products. A persistent identity helps people understand who is participating; it does not prove that every connected action is appropriately scoped. Buyers should ask how identity, workspace access, tool authority, and offboarding remain aligned when a participant or integration changes.

The economic test must include inference and coordination outside the messenger. Our Claude Projects analysis examined how parallel workers consume shared allowance. Ando’s human-seat model does not, by itself, establish the charging terms of an independently connected agent. Ask who pays for its model calls, what is included in hosted execution, and how usage is observed. A predictable seat bill can coexist with a variable operating bill.

Slack is moving, and history does not move itself

The strongest case against switching is not that agents cannot benefit from shared context. It is that an incumbent may supply enough of the needed experience without relocating the company’s communication history. Slack’s August agent update announces multi-agent thread improvements, session controls, and the planned February 2027 deprecation of its older assistant view. The relevant comparison is with that evolving product, not a static image of yesterday’s bot integration.

Control quality is visible in the details. Slack’s Agent Sessions documentation makes its native stop button conditional on the app subscribing to the stop event. The app must then stop its work, clean up, and change the session state. A stop icon is not proof that an underlying action halted. This gives evaluators a concrete acceptance test for either platform: observe the actual worker after cancellation, not only the appearance of the conversation.

Remaining with Slack is not necessarily effortless either. Its migration guide says changing from the assistant view to the agent view cannot be reversed. Distributed apps may need Marketplace review. That is a specific integration change, not a reason to abandon the platform, but it belongs in the incumbent option’s engineering budget. Compare two real paths: upgrading the existing agent experience and moving the relevant workflow elsewhere.

History creates a separate constraint. Slack’s export documentation limits ordinary Free and Pro exports to public-channel messages and file links. Broader exports on higher plans require the documented permissions and approvals. The existence of a downloadable archive therefore does not establish that every private discussion, attachment, or permission relationship can be moved into another product unchanged.

Before promising a migration, identify which historical records the target workflow actually needs. Ask Ando to demonstrate their import or continued accessibility, including who can see them afterward. A team may rationally begin with new project conversations instead of relocating everything. That is a limited pilot, not a completed replacement. The acceptance record should say which platform remains authoritative for old decisions and which one owns new work.

The strongest pro-Ando counterargument is that a new coordination model cannot be judged solely by a feature checklist against an incumbent. Agents that participate naturally may make previously awkward work routine. That is a fair reason to try it. It is not a reason to omit retention, export, cancellation, or cost questions; those are what let a useful experiment survive contact with ordinary operating obligations.

Evidence would change the verdict in either direction. Consistently fewer manual relays, shorter time to accepted output, and manageable correction effort would justify widening the trial. Repeated context disputes, rising model consumption, or a need to keep two fully active communication systems would weaken it. Neither a funding round nor an enthusiastic internal testimonial measures that balance for another team.

Move a workflow before moving the company

A practical trial needs a boundary, an owner, and an exit. Name the work entering the test, the people and agents allowed to participate, the systems they may read, and the actions that remain outside scope. Keep the incumbent workflow recoverable. The purpose is to discover whether the new workspace removes coordination effort, not to demonstrate loyalty to an architectural idea.

Measure accepted output alongside the work people do to obtain it. Record manual forwarding, repeated explanations, correction requests, interrupted runs, and the time needed to establish the current decision. Do not use message volume as a productivity proxy. A more active conversation could mean useful collaboration or extra noise; the distinguishing evidence is whether the underlying job finishes with less total burden.

Today’s Basalt brief values a reported processing-time gap without calling it payroll savings. That is the right discipline for Ando too. Minutes that no longer disappear into context relay may be valuable, but the buyer must observe where they go. Avoid converting every apparent time saving into cash while simultaneously claiming the same employee’s newly available capacity as a second benefit.

The Feedzai brief separates currently offered workflow assistance from upcoming agent-connectivity promises. Ando should face the same current-product test. Require the requested harness, access boundaries, and operating terms to work in the actual pilot account. A roadmap can inform the next review; it should not stand in for a dependency needed to complete today’s task.

Engineering leads: trial Ando where shared agent work already creates manual handoffs. Keep a named reviewer and a clearly bounded project rather than moving every conversation at once.
Finance owners: use $3,480 for forty active annual-billed Pro seats only as a reference. Obtain Ando’s quote, connected-model costs, implementation effort, and the actual avoidable incumbent spend.
Platform owners: test cancellation, access removal, historical records, and tool authority. A convincing conversational interface does not prove that connected work has stopped or that permissions transferred correctly.
Rollout owners: expand when accepted work improves without hidden duplicate workflows. Pause when maintaining context across systems becomes more work than the relay the new platform was meant to remove.

Ando has made a credible product question newly concrete: should an agent be a participant in the place where a team decides what to do? The answer can be yes without making the migration automatic. A shared conversation earns its keep only when it removes a handoff without erasing a boundary.