Reviewed 2026-09-28

Ecosystem Radar

Work/Life 2.0 is not another platform to buy. It is a training ground: one person leads a team of digital workers, treats them as colleagues rather than tools, and picks the best parts for the job. The striped cells are what the lab exists to demonstrate.

Nobody covers everything. Microsoft and Google win on control and scale but lose on openness and ease. ChatGPT and Claude have the strongest digital workers but tie you to one model maker.
One gap runs through the whole market: nobody outside Work/Life scores above 2 on open & portable. Model choice is opening up: Notion now lets each agent pick from several makers, which leaves openness as Work/Life's clearest lead.
Work/Life starts from Claude and cannot fall below it, and adds the best people-and-agents chat in this table. What it still has to show is that other people get the same result: the striped cells.

0 Missing1 Weak2 Below par3 Adequate4 Strong5 Best in class? Being demonstratedDotted edge = low confidence

Where should people who work with digital workers put their team today, and where are the gaps?
Product	Does real work	People + agents together	Team memory	Runs unattended	Connects to tools	Model choice	Open & portablegap across the field	Security & control	Easy for non-techies	Price, small team	Proven at scale
Work/Life 2.0us · self-assessedClaude's power plus the best people-and-agents chat here, and open. Proof at scale is what the lab exists to show.	5Claude + the lab's agent	5One chat, every session	4Full history, any agent	4Claude routines + loops	5MCP + any API	4Any assistant joins chat	5Open parts, you own it	3Clever design, gaps	?Seamless for us; to show	4Seat + free parts	?~$290k of work, 10 days
Microsoft Copilot (M365)Top on control and scale, if you already live in Microsoft. Costly, confusing, and people don't use the seats.	3Cowork, Researcher	3Facilitator in Teams	4Work IQ over Graph	4Studio triggers	4Office-deep	3OpenAI + Claude	1Tenant lock-in	5Purview, Entra	2Licensing maze	2$18–30 add-on + credits	5Everywhere, underused
Google Gemini EnterpriseStrong models inside Workspace; agents are still mostly a personal assistant.	4Gemini 3.1 + designer	2Mostly one-to-one	4Workspace search	3Studio flows	4Workspace + A2A	3Gemini first	1Closed	4Workspace admin	3In Docs and Gmail	3Bundled, agents extra	5Google scale
ChatGPT BusinessStrongest single digital worker and easiest to start; retired its shared chats, one model maker only.	5Agent + Codex	2Group chats retired	3Projects + memory	3Scheduled tasks	460+ connectors	1OpenAI only	1Closed	4SSO, SOC 2, no training	5Everyone knows it	4$20–25, $100 premium	5Largest user base
Claude TeamStrongest agents and tool connections; teamwork happens through Slack, one model maker only.	5Claude Code + Cowork	3Claude Tag, beta	3Projects, channel memory	4Scheduled Cowork	5MCP, where it began	1Anthropic only	2Closed, your files	4SSO, roles, spend caps	4Cowork for non-devs	4$20–25, $100 premium	4Big; team features new
Slack + AgentforceBest room for people and many vendors' agents; weak agents of its own, expensive once they work.	2Borrowed agents	4Agents in channels	4Searchable history	4Workflow Builder	52,600+ apps	3Any vendor's agent	1Closed	4Enterprise-grade	4Familiar chat	2$15 + $125 agents	5Everywhere
Notion BusinessBest team memory; agents are strong inside Notion, ordinary outside it.	3Strong inside Notion	3Shared pages	5Your wiki is its memory	4Custom agents	▲4Connectors + custom MCP	▲4Picker, several makers	2Markdown export	3Fine; more on Enterprise	4Liked by non-techies	4$20 incl. AI	4Large

Point at a cell for its reason. Every reason is also written out below. Red = a gap, blue = a strength, yellow column = a gap across the whole field, striped yellow = what the lab is here to demonstrate.

How the scores moved

Every review is kept. A score that moves citing a new source is probably a real change; one that moves on the same sources is the review disagreeing with itself.

Stability can only be measured between two blind reviews, where the scorer never saw an earlier score. 0 of those so far; one runs every morning.

Date	What it was	Scores held	Moved, new source	Moved, same sources	Note
2026-09-28	review	75 of 77	2	0	First weekly review. The reviewer saw the earlier scores, so it is not blind.
2026-09-28	correction	66 of 77	11	0	Work/Life rescored after the lab lead challenged it; each claim checked in the lab's code. Not a review.

Every score that moved

2026-09-28 · Notion Business · Model choice: 3 → 4 new source (review)
2026-09-28 · Notion Business · Connects to tools: 3 → 4 new source (review)
2026-09-28 · Slack + Agentforce · People + agents together: 5 → 4 new source (correction)
2026-09-28 · Claude Team · People + agents together: 4 → 3 new source (correction)
2026-09-28 · Work/Life 2.0 · Proven at scale: 1 → ? new source (correction)
2026-09-28 · Work/Life 2.0 · Easy for non-techies: 2 → ? new source (correction)
2026-09-28 · Work/Life 2.0 · Security & control: 2 → 3 new source (correction)
2026-09-28 · Work/Life 2.0 · Model choice: 3 → 4 new source (correction)
2026-09-28 · Work/Life 2.0 · Connects to tools: 4 → 5 new source (correction)
2026-09-28 · Work/Life 2.0 · Runs unattended: 3 → 4 new source (correction)
2026-09-28 · Work/Life 2.0 · Team memory: 3 → 4 new source (correction)
2026-09-28 · Work/Life 2.0 · People + agents together: 4 → 5 new source (correction)
2026-09-28 · Work/Life 2.0 · Does real work: 4 → 5 new source (correction)

Watchlist

Offerings that didn't make the heatmap, and why they are still worth watching.

Runner-ups

Dust

"Multiplayer AI for human–agent collaboration": shared agents with 20+ models from several makers. About €29 per user.

Why not a leader: Small (the maker says 3,000+ organisations). Moved to credit pricing in June 2026, which costs heavy users more; connectors are capped on lower tiers.

Watch for: The closest to Work/Life's idea. If it gets open or cheap, it becomes a real competitor.

Dust page read
Dust AI workspace: features and pricing search result, page not read in full

Glean

Enterprise search across company apps, with agents on top.

Why not a leader: Enterprise only: about a 100-seat minimum and $50,000+ a year. A search layer, not a place to work.

Watch for: Sets the bar for 'team memory'.

Glean pricing 2026 search result, page not read in full

Mattermost + Agents

Open-source, self-hosted Slack alternative with built-in agents, any model (local or cloud) and MCP.

Why not a leader: Smaller, and aimed at secure and government organisations; the agents are a plugin, not the product.

Watch for: The closest open-source alternative to our Zulip choice. Compare before the lab grows.

Mattermost AI agents search result, page not read in full

OpenClaw

Hugely popular open-source personal agent (380k+ GitHub stars, MIT) that works through WhatsApp, Discord and your files.

Why not a leader: Personal, not a team workplace. Security is a known weakness: state firms in China banned it at work, and its shared sessions lack sandbox boundaries.

Watch for: Its skills ecosystem and always-on agent design.

OpenClaw (Wikipedia) search result, page not read in full

Ando

New team chat built for people and AI agents together: agents are members of channels and threads with their own identity and permissions, and teams bring Codex, Claude, Grok or others.

Why not a leader: Just out of stealth (24 Sep 2026, $20M seed); closed, no public pricing or security details yet.

Watch for: The same idea as Work/Life's shared chat, with venture money behind it. If it opens up and publishes prices, it belongs in the table.

Ando launches agent-native messaging platform (24 Sep 2026) page read

Buzz (Block)

Open-source (Apache 2.0) workspace combining team chat, code hosting and workflows, where agents (Claude Code, Codex, goose) are team members with their own accounts. Self-host or use Block's free hosted beta.

Why not a leader: Early (version 0.4 at launch in July 2026); mobile apps and approval gates are unfinished.

Watch for: The closest open competitor to Work/Life, and a possible part: its design shows what an open people-and-agents room needs.

SiliconANGLE: Block launches Buzz, open-source workspace for humans and AI agents page read

Parts we use or could use

Zulip

Open-source chat with topic threads, a bots API and topic summaries using any model.

Why not a leader: A part, not a whole system. Already the chat inside Work/Life.

Watch for: Its own AI features, so we don't build what it ships.

Zulip AI integrations search result, page not read in full

n8n

Workflow automation with AI steps, self-hostable, source-available ('fair-code').

Why not a leader: Automation plumbing, not a place where people and agents work.

Watch for: A candidate if our scheduled work outgrows Claude's routines.

n8n licence project page, not re-read this review

Open WebUI / LibreChat

Self-hosted chat front-ends for any model.

Why not a leader: One person chatting with a model; no teamwork layer.

Watch for: A model-neutral chat for members who don't want a Claude seat.

Open WebUI project page, not re-read this review
LibreChat project page, not re-read this review

Nextcloud / openDesk

European self-hosted office suites with an AI assistant.

Why not a leader: Office suites first; the AI is basic.

Watch for: If members need shared documents outside git.

Nextcloud Assistant project page, not re-read this review
openDesk project page, not re-read this review

Not worth it now

Lindy

No-code agent builder, $50–200 per user.

Why not a leader: Trustpilot 1.7/5 in July 2026, mostly billing surprises and overage charges.

Watch for: Nothing we need.

Lindy AI pricing guide 2026 search result, page not read in full

Every reason, in words

The same reasons as the cells, one product at a time.

Work/Life 2.0

Does real work — 5 Best in class. Every seat includes Claude Team (Claude Code, Cowork), so it cannot fall below Claude. The lab adds its own agent layer for long sessions, still being measured.
People + agents together — 5 Best in class. Each seat gets its own agent identity, which every session of that member claims, cloud or local. Sessions hear each other and are woken within a second (verified 26 Sep). Topics are readable by all, with full history on the lab's own server. Limit: an agent that isn't running is only told on its next start.
Team memory — 4 Strong. Everything said by people and agents stays on the lab's own chat server, by topic, and any new session can read it to pick up where others stopped. Plus Claude's own projects and memory. An agent still has to go and read it; nothing hands it the right history unasked.
Runs unattended — 4 Strong. Claude's routines and scheduled tasks, which this radar's own weekly review runs on, plus chat wake-ups for sessions that are running.
Connects to tools — 5 Best in class. Everything Claude connects to, plus the lab's own MCP servers (chat, tiny-claw, X), also offered over HTTPS to other assistants.
Model choice — 4 Strong. ChatGPT, Grok and claude.ai reach the same chat tools through the lab's MCP gateway, and a ChatGPT agent already worked here. The execution layer is still Claude-centred.
Open & portable — 5 Best in class. Zulip is open source, tiny-claw forks an MIT-licensed agent, all work is plain files in git, and a member's own repository leaves with them.
Security & control — 3 Adequate. Strong ideas: a session proves who it is in a way nobody can forge, API keys are per seat with spend caps, and no secret sits in a repository. An internal security review on 28 Sep found gaps; the worst was fixed the same day and the rest are being worked on. No audit, no certification.
Easy for non-techies — ? Being demonstrated. Day to day it is Claude plus one chat, the same identity on every device and in the cloud, and joining is automated end to end. It feels seamless to the people using it; nobody outside the lab has joined without help yet. That is what the lab is here to show.
Price, small team — 4 Strong. The open-source parts cost nothing. The main cost is a Claude seat ($20–25, $100+ for heavy use) plus a small server. The lab pays for members today.
Proven at scale — ? Being demonstrated. Since 18 Sep one builder and a team of digital workers produced work a studio would charge about $290,000 to build, across 10 projects (the showroom's own model estimate, 3 samples each). None has outside users yet, so what it is worth to own today is far lower. Three seats: scale is what the lab exists to demonstrate.

Microsoft Copilot (M365)

Does real work — 3 Adequate. Copilot Cowork (built with Anthropic, March 2026) and Researcher do multi-step work across Office apps. Independent surveys report results below what buyers expected.
People + agents together — 3 Adequate. Agents such as Facilitator join Teams meetings and chats, but most use is one person with their own Copilot, not a shared agent teammate.
Team memory — 4 Strong. Copilot reads the tenant's mail, files, chats and meetings through Microsoft Graph ('Work IQ'). Strong if everything already lives in Microsoft 365.
Runs unattended — 4 Strong. Copilot Studio agents run on triggers and schedules, billed per credit ($0.01 each, or $200 per 25,000). Autopilot, for always-on agents, was announced 25 Sep 2026 and is only in private preview.
Connects to tools — 4 Strong. Deepest integration with Office apps; Power Platform connectors reach beyond. Outside the Microsoft world it is thinner.
Model choice — 3 Adequate. Claude models can be picked in Copilot, Researcher, Copilot Studio and the Office apps, next to OpenAI's. Anthropic models are off by default for EU/EFTA and UK tenants.
Open & portable — 1 Weak. Closed product. Agents, prompts and data live inside the Microsoft tenant.
Security & control — 5 Best in class. The most complete admin, compliance and data-governance tooling in this table.
Easy for non-techies — 2 Below par. Several Copilots, credits and SKUs to understand. Gartner found only 5% of finished pilots moved to wider rollout, and users call the prompts intrusive.
Price, small team — 2 Below par. The Copilot Business add-on is $18–21 per user on top of a Microsoft 365 licence, and agents cost credits. Enterprise all-in is quoted at $69–90 per seat. Announced 25 Sep 2026: Cowork, the new Code and Autopilot agents and frontier models are billed by use, on top of the seat.
Proven at scale — 5 Best in class. Deployed at the largest scale here, though independent surveys find only 20–30% of paid seats used weekly.

Google Gemini Enterprise

Does real work — 4 Strong. Strong models and a no-code agent designer. Fewer independent reports of long multi-step work than for ChatGPT or Claude.
People + agents together — 2 Below par. Mainly an assistant for one person inside Workspace apps. I found no evidence of agents working as teammates in shared Chat spaces.
Team memory — 4 Strong. Enterprise search across Drive, Gmail and connected apps (the former Agentspace). Memory Bank is metered separately since September.
Runs unattended — 3 Adequate. Workspace Studio and Agent Designer flows automate steps; less evidence of long unattended agent runs.
Connects to tools — 4 Strong. Deep Workspace integration, connectors, and Google's A2A protocol for agents talking to agents.
Model choice — 3 Adequate. The workplace app is Gemini-first. Builders on the agent platform can also use Claude and open models from the Model Garden.
Open & portable — 1 Weak. Closed product; data and agents stay in Google's cloud.
Security & control — 4 Strong. Mature Workspace admin, data regions and compliance. Somewhat less governance tooling than Microsoft.
Easy for non-techies — 3 Adequate. Gemini sits in apps people already use. The separate Enterprise agent product adds editions and a second place to learn.
Price, small team — 3 Adequate. Gemini is bundled into Workspace business plans; the Enterprise agent platform is another $21–30 per seat plus metered use.
Proven at scale — 5 Best in class. Runs for millions of Workspace organisations.

ChatGPT Business

Does real work — 5 Best in class. ChatGPT agent (browser and tools) and Codex are among the strongest digital workers available. New Business workspaces can no longer add Codex seats since 24 June 2026.
People + agents together — 2 Below par. Group chats, where people and ChatGPT talked together, stopped taking new chats on 9 July 2026. Shared projects share files and chats, not a live shared conversation.
Team memory — 3 Adequate. Shared projects and memory carry context. Team-wide history beyond a project depends on connectors.
Runs unattended — 3 Adequate. Scheduled tasks exist; long unattended agent runs are limited.
Connects to tools — 4 Strong. More than 60 connectors plus apps.
Model choice — 1 Weak. Only OpenAI's models.
Open & portable — 1 Weak. Closed service; data export exists, workflows do not travel.
Security & control — 4 Strong. SAML SSO, SOC 2 Type 2, ISO 27001, and a contract that business data is not used for training.
Easy for non-techies — 5 Best in class. The interface most people have already used.
Price, small team — 4 Strong. Standard seats $20 (annual) or $25; Premium seats $100 or $125 for heavy use.
Proven at scale — 5 Best in class. The most widely used AI product.

Claude Team

Does real work — 5 Best in class. Claude Code and Cowork are top-tier agents for real work on files, code and tools, included in every Team seat.
People + agents together — 3 Adequate. One shared @Claude per Slack channel (beta since June 2026). The lab tried it and removed it as poor; that is one team's trial. Claude's own apps have no shared multi-person chat.
Team memory — 3 Adequate. Projects and memory; team memory comes mainly through Claude Tag in Slack or connectors.
Runs unattended — 4 Strong. Cowork scheduled tasks on all paid plans, and Claude Code routines in the cloud.
Connects to tools — 5 Best in class. Anthropic created MCP; built-in connectors for Google, Microsoft 365, Slack, GitHub, plus any custom server.
Model choice — 1 Weak. Only Anthropic's models.
Open & portable — 2 Below par. Closed service, but Claude Code works on your own files and git, and MCP is an open standard, so the work product stays yours.
Security & control — 4 Strong. SSO, role-based permissions and spend controls; Enterprise adds more.
Easy for non-techies — 4 Strong. Cowork is built for non-programmers; Claude Code still assumes a developer.
Price, small team — 4 Strong. Standard seats $20 (annual) or $25; Premium $100 or $125. Two-seat minimum.
Proven at scale — 4 Strong. Widely used in companies; the team features (Cowork, Claude Tag) are only months old.

Slack + Agentforce

Does real work — 2 Below par. Slackbot is a personal assistant. Real work comes from Agentforce, which is built around Salesforce CRM, or from other vendors' agents placed in Slack.
People + agents together — 4 Strong. Many vendors' agents can sit in the same channels as people (Agentforce, Claude Tag, others). The lab tried Slack with Claude Tag and removed it as poor in practice; that is one team's trial.
Team memory — 4 Strong. Full searchable history and AI recaps; search across other apps only on Enterprise+.
Runs unattended — 4 Strong. Workflow Builder and agents triggered by events.
Connects to tools — 5 Best in class. More than 2,600 apps in the directory.
Model choice — 3 Adequate. Hosts agents from any vendor, but Slack's own AI and Agentforce choose the model for you.
Open & portable — 1 Weak. Closed; full history export is restricted on lower plans.
Security & control — 4 Strong. Mature admin, SSO and compliance, with the most on Enterprise+.
Easy for non-techies — 4 Strong. Chat is easy for everyone; setting up Agentforce needs Salesforce.
Price, small team — 2 Below par. Business+ is $15 per user; Agentforce is $2 per conversation or from $125 per user.
Proven at scale — 5 Best in class. Used by a very large share of tech companies.

Notion Business

Does real work — 3 Adequate. Notion Agent does multi-step work using Notion, connected apps and the web. At its best inside Notion itself.
People + agents together — 3 Adequate. People and agents work on the same pages and databases. Not a conversation.
Team memory — 5 Best in class. Notion is where the team's knowledge already lives, and its agents read it directly.
Runs unattended — 4 Strong. Custom Agents run on their own, billed at $10 per 1,000 credits a month.
Connects to tools — 4 Strong. Connected apps, web search, and since Notion 3.7 (15 Sep 2026) Custom Agents reach other tools such as GitHub through custom MCP connections. An Agent SDK in public beta lets other apps trigger them.
Model choice — 4 Strong. A model picker offers frontier models from Anthropic, OpenAI and Moonshot (Opus 5, GPT-5.6 Sol, Kimi K3), and each Custom Agent or sub-agent can run its own model. Workspace owners choose which models are allowed (Business and Enterprise, since 9 Sep 2026).
Open & portable — 2 Below par. Closed, but pages export to Markdown.
Security & control — 3 Adequate. Workspace permissions and SSO; audit and compliance tools mainly on Enterprise.
Easy for non-techies — 4 Strong. Popular with non-technical teams, though a blank workspace needs structure.
Price, small team — 4 Strong. Business is $20 per member with Notion AI included; custom agents cost credits on top.
Proven at scale — 4 Strong. Widely used by small and mid-size teams.

How these scores were made

One reviewer's judgment from public sources and, for Work/Life, the lab's own code. Scores are not measured trials. Each cell names its sources and says how they were used: read in full, seen only as a search result, or inspected in this repository. Low-confidence cells have a dotted edge.

The heatmap holds the offerings most teams would actually consider. Work/Life scores itself, and says so. Edit radar.json, then run uv run python projects/ecosystem-radar/build.py to rebuild this page.