BlackOak.agency// AGENTIC COMPANIES · THE LANDSCAPE
Anyone setting up an agentic organisation gets dozens of tools pitched at them. This is the overview we use with clients: what each category does, what the best-known names can and cannot do, and why our default choice is AI-Office and Paperclip.
Research updated on September 11, 2026
// OUR CHOICE
We are not married to a tool. We looked at Multica, Gas Town, Cofounder and the enterprise platforms, and we run several of them ourselves. Five reasons our default still lands on these two.
Both run on your server, in the EU if you want, with their own Postgres database. Paperclip is MIT; with AI-Office you get the source. No per-seat licence, no vendor that can pull the plug.
Paperclip is the only open-source platform that ships org chart, delegation, heartbeats, monthly budget with hard stop and approvals in one package. Multica lacks budgets and hierarchy, Gas Town and Symphony only do code, Cofounder is closed.
AI-Office is deliberately small: a few scheduled agents with MCP connections, skills and an inbox. That gets the first agent live within two weeks. When it grows into agents managing each other, we port the design one to one to Paperclip.
Paperclip runs Claude Code, Codex, Gemini CLI, Cursor, OpenClaw or any HTTP agent. Frontier is OpenAI-only, Managed Agents is Claude-only, Cofounder accepts no bring-your-own model keys. Who is best or cheapest changes every quarter; you want to be able to switch.
Our AI-Office watches our sites, reviews our code and writes our briefs, every day. We know the pitfalls of scheduled agents from experience, not from a brochure. That knowledge comes with us into your setup.
A software team that wants agents in its backlog: Multica, possibly next to Paperclip. An organisation fully on Microsoft 365 that mainly wants control over agents staff already use: Agent 365 as a governance layer. A process with fixed steps and one AI step: n8n, and we say honestly that this is rung 2. An agent that has to live inside your product: Claude Agent SDK or CrewAI, no control plane.
// THE CATEGORIES
Self-hosted platforms with a coordination layer above the agents. The only group that really competes with Paperclip.
Issue tracker where you assign an issue to an agent like to a colleague. Supports 26 coding CLIs, cron autopilots, execution logs and token cost per issue. Docker Compose or Helm; a cloud edition too.
Fits: software teams that want agents as team members in their backlog.
Lacks: org chart, delegation between agents, budget caps, approval before execution.
“Factory” for 20 to 30 parallel coding agents with fixed roles (Mayor, Polecats, Refinery, Witness), git-native with a merge queue. Gas City is the declarative successor.
Fits: one experienced operator running a large fleet of coding agents.
Lacks: anything outside software development; no budgets or business roles; steep learning curve.
Turns issue-tracker tickets into isolated, autonomous Codex runs. OpenAI calls it a spec they deliberately won’t productise.
Fits: teams already on Codex wanting the lightest form of “tickets in, PRs out”.
Lacks: agents other than Codex, org chart, budgets, approvals.
A manager agent directs workers (OpenClaw, QwenPaw, Hermes) in a Matrix chat room where humans watch and intervene. Since August 2026 audited interventions and project history.
Fits: organisations that want to see coordination in a chat room, with one layer of hierarchy.
Lacks: budgets, deeper org charts; smaller, mostly Chinese community.
Established self-hosted platform for autonomous coding agents in sandboxes, headless and reliable.
Fits: teams that mainly need a robust, headless coding agent.
Lacks: organisational model and budgets.
Projects copying the Paperclip idea: named agents on an org chart, escalation to a human, departments delegating to each other.
Fits: reading as a reference for design choices.
Lacks: community and maturity; not something to recommend to a client.
The same idea as Paperclip, but as closed SaaS. Quick to start, no ownership.
A company of agents in departments (engineering support, sales, marketing, design, support, ops, finance, legal) under a coordinator agent, with shared context. Approval for consequential actions, daily, weekly or monthly routines, MCP and managed integrations with GitHub, Vercel and Supabase. Aimed at founders launching a product.
Fits: solo founders and small startups wanting a “company in a box” without engineering.
Lacks: self-hosting, open source, bring-your-own model keys, per-agent budgets; open beta, so young and changing.
End-to-end platform to build, deploy and manage agents “like employees”: onboarding, identities, permissions, feedback loops and shared business context. Launch customers HP, Intuit, Oracle, State Farm, Thermo Fisher, Uber.
Fits: large enterprises with OpenAI as their standing vendor.
Lacks: self-hosting, other models, an open org chart; outside the SME profile.
Control plane for observing, securing and governing agents inside and outside Microsoft, including OpenClaw and Claude Code on workstations and imports from AWS and Gemini. Building happens in Copilot Studio or the Agent Framework.
Fits: organisations fully on Microsoft 365 that want control over existing agents.
Lacks: orchestration of work, per-agent budgets, org chart; it is governance, not operations.
APIs for sandboxing, long sessions, scoped permissions, tool execution and tracing. Infrastructure for one agent, no organisational layer.
Fits: teams wanting to run one Claude agent without their own infrastructure.
Lacks: other models, data residency (runs at Anthropic), org chart and budgets.
Agents tied to the vendor’s platform. Building blocks with governance, no open organisational model.
Fits: organisations for which that platform is already the system of record.
Lacks: portability; you build in the vendor’s cloud and stay there.
Sell “AI employees” to non-technical teams. Fast results, but you rent and own nothing.
Visual builder for agent teams, strong in sales and GTM, 2,000+ integrations.
Fits: sales and marketing teams.
Lacks: self-hosting, ownership, hard-stop budgets.
“AI employee” in plain language, 5,000+ integrations, Agent Swarms, phone AI.
Fits: professionals who want email, calendar and phone handled.
Lacks: self-hosting, ownership, org chart.
Enterprise platform for document-heavy back-office processes with deterministic, auditable outcomes.
Fits: larger organisations with heavy document processes.
Lacks: a lightweight start; it is an enterprise programme.
Build one agent or one collaboration pattern in code. Not a replacement for a control plane, but what you use when the agent has to live inside your product.
Role-based agent teams; closest in spirit to Paperclip, but you write the orchestration yourself.
Fits: developers building a fixed collaboration pattern into a product.
Lacks: org chart, budgets, approvals; you build those on top.
State machines with checkpointing; fine-grained control at a lower level than an org chart.
Fits: complex, deterministic flows inside your own application.
Lacks: everything organisational; it is a library.
The big labs’ SDKs, each with a natural path into its own cloud. AgentScope is the base under AgentTeams.
Fits: teams already building in that ecosystem.
Lacks: neutrality; gravity pulls towards Azure, OpenAI or Google Cloud.
What our AI-Office uses under the hood, respectively Vercel’s agent framework with durable sessions and schedules.
Fits: an agent inside a Next.js or Node product.
Lacks: an organisational layer; that is AI-Office or Paperclip.
Fixed steps with an AI step in between. Excellent for rung 2 on our ladder; no autonomous agents with their own goals.
Workflow automation with 400+ integrations and AI nodes.
Fits: processes with a fixed route and one or two AI steps.
Lacks: agents choosing their own route, org chart, budgets.
Visual LLM builders with RAG and agents; Flowise and Langflow are now owned by Workday and IBM.
Fits: quickly an assistant with your own documents.
Lacks: organisational model and oversight of autonomous agents.
Not an alternative to a control plane, but the layer beneath it: the agent doing the work, driven from Paperclip, Multica or AgentTeams.
Personal agent in your messaging apps (Telegram, WhatsApp, Discord, Signal), with memory and tools.
Fits: as the employee under Paperclip; “if OpenClaw is the employee, Paperclip is the company”.
Lacks: coordination of multiple agents.
Self-improving agent with persistent memory, 40+ tools and model routing.
Fits: likewise, as a runtime under a control plane.
Lacks: organisational layer.
// SIDE BY SIDE
Paperclip sets the bar: org chart, heartbeats, budget with a hard stop, approvals, bring your own agent, self-hosted. This is how the direct competitors score on the same points.
| Criterion | Paperclip | BlackOak AI-Office | Multica | Gas Town | Cofounder | Agent 365 | OpenAI Frontier |
|---|---|---|---|---|---|---|---|
| Org chart & delegation | Yes | Flat, handoff | No | Fixed roles | Departments | No | Identities |
| Heartbeat / schedule | Yes | Cron | Autopilots | Continuous | Routines | n/a | Yes |
| Budget with hard stop | Yes | Per run | Insight only | No | No | Policy | Unknown |
| Approval before execution | Board | Inbox | Via PR | No | Yes | Policy | Yes |
| Bring your own agent / model | 7+ | Claude | 26 CLIs | Claude, Codex, Gemini | No | Discovery | OpenAI |
| Self-hosted, own data | Yes | Yes | Yes + cloud | Yes + Kilo | Cloud | Cloud | Cloud |
| Beyond software development | Yes | Yes | Code | Code | Yes | Yes | Yes |
| Licence / price | MIT | Source from us | Apache 2.0+ | MIT | $120/mo | $15/user/mo | Enterprise |
| Maturity | 6 mo, very active | Ours, in use | 8 mo, very active | 9 mo, v1.0 | Open beta | GA | GA |
Every open-source option comes with hosting, near-daily updates, backups, key management and the audit log the EU AI Act requires for high-risk systems from December 2027. We price that in.
// ADVICE
Several agents working together, self-hosted, beyond code
Paperclip. No other open-source project ships org chart, budget and approvals in one package. We host and monitor it.
Start small with a few monitoring or delivering agents
Our AI-Office. First agent within two weeks; move to Paperclip once agents need to hand work to each other.
Software team wanting agents as team members in the backlog
Multica, possibly next to Paperclip. Assign an issue, get a PR back, cost per issue.
Solo founder wanting a quick “company in a box” without engineering
Cofounder is a fair option then, as long as you accept owning nothing and the product being young. We prefer helping with a base you own.
Fully on Microsoft 365, control over agents staff already use
Agent 365 as the governance layer; we arrange orchestration of work separately.
The agent has to live inside the product
Claude Agent SDK, CrewAI or LangGraph. No control plane needed.
Fixed steps with one AI step
n8n. And we say honestly this is rung 2, not an agentic organisation.
Want to know which of these tools fits your organisation? We walk through it with you in half an hour, no strings attached.
Book a call →