Governed enterprise AI
Delegate real work to AI with policy, approvals, and audit trails built in from day one.
Goal
Review 3 vendor contracts due for renewal and recommend action
of companies expect to use agentic AI at least moderately by 2027
of companies expect to customize AI agents for their unique business needs
of organizations still use AI at a surface level with little process change
THE GAP
Enterprises have invested heavily in AI — chatbots, copilots, and RPA tools. These tools are useful, but many still depend on people to carry work across systems, approvals, and follow-up.
Surface-level assistance without authority.
→ Manual follow-up slows every process
Requires constant human steering.
→ Hidden automation is hard to trust
Brittle workflows that break on change.
→ Rigid automation breaks frequently
Static routes lacking dynamic judgment.
→ General agents lack enterprise controls
The concept
A Softworker accepts a goal in plain language, breaks it into a plan, uses only approved tools, asks for human sign-off when it matters, and leaves a full audit trail behind.
Describe the desired outcome in plain language. Softworkers accept high-level business goals (e.g. "Prepare renewal packets") instead of requiring fragile, step-by-step prompts.
The worker decomposes the goal into tasks, identifying necessary tools, access boundaries, data inputs, and organizational policies before any execution starts.
For high-risk systems or actions above defined thresholds, the Softworker pauses execution. It presents findings and waits for explicit approval from an authorized human.
Every action, document analyzed, decision, and approval is logged in an immutable audit trail. This ensures total transparency and compliance for security and operations.
Less prompting. More governed work.
Goal
Human delegates work in plain language
Plan
Softworker breaks it into governed steps
Approval
Sensitive actions wait for human sign-off
Audit
Every action is logged and reconstructable
Beliefs
Six commitments that are non-negotiable in everything we build - the principles that separate governed execution from blind automation.
No anonymous agents. Every Softworker has an assigned owner, a defined access scope, and a logged identity on every action.
Every connector is governed. Nothing is open by default.
Sensitive thresholds are defined upfront. The worker waits.
Humans are always in control, not just at the start.
Full audit trail. No black boxes. Ever.
Guardrails are set at the organisational level, not left to individual workers or users. The system enforces what matters.
Early access
We're working directly with a small group of enterprise teams to shape what governed AI execution looks like in practice. If the problem resonates, we'll get on a call within a week.
We read every submission. If it's a fit, you'll hear from us directly. No automated sequences.
A Softworker is an autonomous AI worker you deploy inside your organisation. You delegate a goal, such as renewing vendor contracts or processing expense reports, and the Softworker plans the steps, connects to the systems it needs, requests human approvals where required, executes the approved actions, and returns an auditable record. It is not a chatbot you prompt repeatedly. It is a worker you assign tasks to.
Chatbots answer questions. Copilots suggest next actions and wait for you to act. Softworkers actually move work forward autonomously. The difference is execution authority: a Softworker can draft a contract, route it for approval, send it via DocuSign, and log the outcome without a human steering each step. Copilots surface information. Softworkers complete work.
No. RPA automates rigid, pre-scripted click sequences and breaks the moment a UI or process changes. Workflow tools like Zapier or Make route data between apps on fixed triggers. Softworkers are goal-driven: they reason about how to accomplish an objective, adapt when conditions change, coordinate across multiple systems, and escalate to humans when they reach a decision point they are not authorised to make alone.
General AI agents are powerful but ungoverned. They can take arbitrary actions, connect to any system, and are hard to audit. Softworkers are AI agents with enterprise governance built in from the start: identity-bound, policy-constrained, approval-gated, and fully auditable. The governed part is not a feature add-on. It is the architecture.
Only within the boundaries you define. When you deploy a Softworker, you specify which actions it can perform autonomously and which require a human approval step. High-stakes actions such as sending contracts, initiating payments, or deleting records can be gated behind mandatory approval flows. The Softworker pauses, surfaces the context to the right person, and waits for sign-off before proceeding.
Through policies. You define the integrations a Softworker is permitted to use, the data it can read and write, the actions it can take autonomously, and the conditions under which it must escalate. Policies are set by your team and enforced at runtime. The Softworker cannot act outside them regardless of the goal it is given.
Every action is logged with full context: what goal was set, what plan was generated, which steps were approved, what was executed, and what the outcome was. This gives you a complete audit trail to understand exactly what happened and why. For high-risk workflows, we recommend requiring explicit human approval before irreversible actions, which eliminates the category of silent mistakes entirely.
When a Softworker reaches a step that requires human sign-off, it surfaces a structured approval request including the goal, the proposed action, the supporting evidence it gathered, and the downstream impact. The approver can approve, reject, or redirect with a note. The Softworker resumes from that decision point with the outcome recorded in the audit log.
Your data is processed only within the systems and boundaries you configure. We do not train foundation models on your data. Data accessed by a Softworker during a task is used only for that task and is subject to the retention and security controls you set. We are happy to discuss data residency requirements during the early-access programme.
Each Softworker operates under a defined identity with scoped credentials. It connects to your systems using OAuth or API keys you provision, with the minimum permissions necessary. No Softworker has standing admin access. Access is granted per task and revocable at any time. Your existing IAM policies remain the source of truth.
Yes. Integration scope is defined at deployment time. A Softworker handling vendor renewals can be restricted to your procurement system and DocuSign and cannot reach your HR or finance systems unless you explicitly permit it. Scope is enforced at the platform level, not just by prompt instruction.
Submit the form on this page. We review every application personally with no automated sequences. If your problem resonates with what we are building, we will get on a direct call with your team within a week. Early-access participants work closely with our founders to shape the product roadmap.
For the initial early-access phase, we work directly with your team to configure and deploy. Our goal is to make Softworkers deployable by operations and business leaders without requiring deep engineering involvement, but we are not there yet for all integrations. This is one of the things we are actively building with early-access partners.
Operations, finance, procurement, and HR tend to see the fastest impact. These functions run high-volume, process-heavy work that is currently coordinated manually across email, spreadsheets, and disconnected tools. Sales operations, legal, and IT have also shown strong fit. The common denominator is work that requires judgment and coordination across systems, not just data retrieval.
We are in early access and are not charging during this phase. Early-access partners get in at no cost in exchange for close collaboration, feedback, and helping us understand real enterprise workflows. Commercial pricing will be announced before general availability.
We are in a focused early-access programme with a small number of enterprise teams. The platform is real and functional, but it is evolving rapidly. Early-access partners should expect rough edges, direct access to our team for support, and a say in what gets built next. We do not recommend deploying Softworkers in critical paths without close collaboration with us at this stage.
Softworker
