Slack Is Not Ready for Agents — Ando vs QoderWake vs Bot in Channel
Bolting a chatbot onto Slack is not agent staffing. Pick a staffing model — desk agent, chat-native platform, or agent-native messenger — then put a stop line on every channel.
Bottom line: Bolting a chatbot onto Slack is not agent staffing. Ando just came out of stealth with a $20M seed and an explicit bet that Slack-class tools were not designed for agents as first-class members. QoderWake already ships digital employees into DingTalk, Feishu, and WeCom — the chat apps where a huge share of Asia-Pacific work actually happens. Grok Bot still wins when you own a cloud desk. If your plan is "invite a bot to #eng and hope," you are buying notification spam with liability attached. Pick a staffing model: desk agent, chat-native agent platform, or agent-native messenger — then put a stop line on every channel the agent can speak in.
This is the chat sequel to Getting Started with QoderWake and sits next to The Age of Proactive AI.
The problem Slack was not built to solve
Classic team messengers were built for humans:
- Identity = person + email + company login group
- Permissions = channel membership and workspace roles
- Memory = scrollback humans skim
- Authority = whatever a human typed
Agents need different building blocks:
- A stable role identity with limited tools
- Task memory that lasts across threads
- Permission to act (not only chat) with an approval gate
- The right to join, summarize, and escalate without being @-mentioned every time — only when policy allows
Sara Du's Ando pitch, as TechCrunch reported it, is blunt: customers kept asking how to run agents inside Slack, and the token, context, and identity hurdles made a new messenger rational. You do not have to leave Slack tomorrow to accept the diagnosis. You do have to stop pretending a workflow builder and a custom emoji are a digital employee.
Three staffing models — choose before you buy
| Step | Desk agent 💻 | Chat-native 💬 | Agent-native messenger 🏠 |
|---|---|---|---|
| 1 | 👤 Human opens the agent app | 👥 Team already in group chat | 🤖 Agents are members like people |
| 2 | 🖥️ Shared computer / cloud desk | @agent routes to named roles | Own inbox + identity |
| 3 | ✅ Approval before send or spend | 👤 Human gates in the thread | Can join and message when policy allows |
| Best when | You own the queue; work spans browser, inbox, CRM, code | Decisions already live in the thread | You will change the home app so agents are members, not plugins |
In plain English: pick where the work lives — your desk, today's chat app, or a new messenger built for agents — then staff that one surface.
| Model | Poster child | Main risk |
|---|---|---|
| Desk agent | Grok Bot (Cursor / xAI) | Team never sees the work; shadow IT of personal agents |
| Chat-native runtime | QoderWake Wakers | Vague roles → mushy routing; automating before a dry-run |
| Agent-native messenger | Ando | Migration cost; early-product governance still on you |
| Bot in channel 🚫 | Slack/Teams app bolted on | Thin context, wasted spend, false sense of staffing |
"Bot in channel" is listed so you can reject it on purpose.
Ando — what the launch actually claims
From TechCrunch and Ando's own launch materials (September 24, 2026):
- Team messaging built so humans and agents share channels, DMs, group chats, and live calls ("Jams") with transcripts agents can use
- Agents get identity and inboxes, can browse/join channels, and can message humans proactively when policy allows
- Agent-agnostic: bring Claude, Codex, Grokbot, or other connectors — Ando is the room, not the only brain
- Backing: Accel, Index, Emergence ($20M seed reported)
- Early customers skew small teams across software, real estate, finance; Ando's own company uses it daily
Founder framing (paraphrased from public quotes, not invented): Slack can add agents; starting from scratch lets you design for what agents are capable of rather than what old scaffolding allows.
Operator translation: Ando is a bet that permissions + context + membership beat plugin theater. It does not replace your stop-line policy. A proactive agent that can DM your CFO still needs Propose+wait on spend.
QoderWake — chat-native without leaving Asia's stack
I covered install-to-first-week in Getting Started with QoderWake. The strategic point for this post:
- Wakers are named digital employees with skills, connectors, memory, and workspaces
- @Waker puts them in the group chat your org already uses
- Groups (Leader + complementary roles) and WakerFlow give handoffs and human gates
- Best default: org work that already lives in the thread
If your company runs on Feishu or DingTalk, QoderWake is not a science project — it is closer to how work already feels. If you are Slack-only in the US, Ando is the clearer "replace the room" narrative; QoderWake is still the clearer "staff the existing room" narrative in Asia-Pacific.
Grok Bot — still the desk default
When the work is "research → draft → wait for my send" across tools you control, Grok Bot remains the staffing model I use daily. Multiple named Bots, shared computer, Auto Review on sensitive actions. It is a poor substitute for a war-room thread with eight humans and two vendors. Different home, same stop-line discipline.
Decision table for CEOs and Heads of Ops
| Question | If yes → lean | If no → lean |
|---|---|---|
| Does the critical work already happen in group chat? | QoderWake or Ando | Grok Bot / desk agents |
| Will you change the company messenger this year? | Ando (or wait) | QoderWake on current chat, or desk agents |
| Do you need agents as members with inboxes? | Ando | @Waker / bot with strict routing |
| Is your risk posture "no autonomous external send"? | Any — with Prepare-only + logged approval | Do not buy proactive join yet |
| Are most teams in Asia-Pacific chat stacks? | QoderWake first | Evaluate Ando / Slack roadmap honestly |
The anti-pattern checklist — Bot in Channel
Kill these before they become "our AI strategy":
- One mega-bot that answers HR, eng, and finance — no role, no memory hygiene
- Read-all-channels permissions "so it has context"
- No dry-run before autonomous or scheduled modes
- Secrets in thread ("here's the API key, bot")
- Customer channels with agents before Prepare-only is proven internally
- No owner for what the bot is allowed to say externally
- Measuring success as message count instead of closed tasks with real outputs
If three or more are true, you do not have chat-native staffing. You have a liability with a webhook.
A 21-day pilot that respects reality
| Days | Move | Success signal |
|---|---|---|
| 1–3 | Pick one staffing model; write the stop-line for that surface | One-pager signed by CEO/CTO |
| 4–7 | Stand up one role (e.g. "Release blocker summarizer") on Prepare-only | Output quality reviewed by a human lead |
| 8–14 | Put it in a test channel or Ando workspace you can afford to break | Routing works; no secret leakage; approvals logged |
| 15–21 | One Propose+wait path (e.g. draft customer reply, human sends) | Zero unauthorized external sends |
Do not pilot three messengers and four agent vendors in the same sprint. That is how you get a slide, not a staffing system.
How this ties to capital and trust elsewhere
The same week Ando launched, Instinct's $1B raise showed consumer agents are a funded land grab, Muse kept pressing the trust question, and OpenAI's agent leak disclosures reminded everyone that containment is unfinished even at the labs. Enterprise chat is not insulated from that weather. If anything, a bot with channel history is a denser trust problem than a single-user consumer agent.
Govern the room like you govern production access. Because for most companies, the room is where production decisions get made.
Bottom line, again
Slack can add agents. Teams can add agents. That is not the interesting sentence.
The interesting sentence: agents need membership, memory, and a stop line — or they need to stay on a desk you supervise.
Ando is the new-messenger bet. QoderWake is the chat-native digital-employee bet for stacks that already live in chat. Grok Bot is the desk bet. Bot-in-channel is the demo bet.
Pick one. Pilot for 21 days. Promote only what survives Prepare-only. Then talk about rolling it to customer-facing channels — not before.
References
- Ando wants to take on Slack… — TechCrunch
- Ando docs
- QoderWake product
- Viral AI agent Instinct raises $1B… — TechCrunch
- Meta debuts Muse. Will consumers trust it? — TechCrunch
- Unsecured OpenAI agents posted 53 user images… — TechCrunch
- Related: Getting Started with QoderWake
- Related: The Age of Proactive AI
- Related: The Agent Stop Line Playbook
CTA: Want this framed for your leadership offsite? Speaking & private briefings. Prefer a written pulse check first: CEO Scorecard. Ongoing operator notes: blog. Show hosts covering People-AI and production agents: podcast guest.
Enjoyed this? Let's work together.
I help companies turn AI strategy into shipped, revenue-generating products.