ARK is the product and research ecosystem developed by ARC Transformation Group.
Infrastructure for the agent-operated enterprise.
Software is starting to act. Agents hold credentials, call tools and change real systems without a person watching each step. ARK is being built for the part nobody solved: whether those actions can be trusted, verified and governed.
Two questions
ARC and ARK are not the same thing.
- ARC Transformation Group
- How should we use technology?
- The company. The commercial relationship, and the advisory and delivery work across cloud, data, security, software and AI.
- ARK
- How should intelligent software interact with technology?
- The technology ecosystem. Products, protocols and research for the layer where software acts on its own behalf.
The system
- Measure
- Verify
- Govern
- Optimize
Evidence, not claims.
Measure what happened. Verify it against the system that was supposed to change. Govern the authority it ran under. Optimize against cost per verified outcome rather than cost per token.
Agent operating lifecycle
- 01
Discover
Find out which capabilities, tools and resources exist, and which ones apply to the task in hand.
- 02
Authenticate
Prove which agent is acting, on whose behalf, and under which credential.
- 03
Understand
Read schemas, contracts and side effects well enough to call a tool correctly the first time.
- 04
Permission
Resolve what this identity is allowed to do right now, including the actions that need a human.
- 05
Execute
Run the work in an environment that fits its privacy, latency, policy and cost constraints.
- 06
Verify
Collect evidence from the systems that were supposed to change, not from the agent's own report.
- 07
Measure
Attribute cost, latency and success to a unit of work a business can recognise.
- 08
Remember
Persist state so the next run starts from what happened, not from an empty context.
- 09
Audit
Keep a record someone can defend later: what was attempted, under what authority, with what result.
The execution gap
Your agent said it finished. Did it?
A successful model response is not the same thing as a successful business outcome.
- An agent can complete a run while the payment failed.
- It can report that a record was updated when the destination system never changed.
- It can call the correct tool with the wrong authority.
- It can finish a workflow while violating the policy that should have governed it.
What the run reported
- Status
- completed
- Steps
- 7 / 7
- Tool calls
- 12 ok, 0 failed
- Latency
- 4.2s
What assurance asks
- What did the agent attempt?
- What authority did it use?
- What actually changed?
- What evidence exists?
- Was the outcome verified?
- What did the verified outcome cost?
Execution success is not a verified outcome.
Products
Four products. Each with its real maturity attached.
The runtime work, the protocol work and the individual scanners are deliberately not on this list. They sit underneath, in Developers and Labs, because presenting an experiment as a product is how trust gets spent.
ARK Control
Outcome assurance for production AI agents.
Understand what agents attempted, what authority they used, and whether the intended outcome actually occurred.
Where this is today: ARK Control is in design-partner development. It is not a product you can buy or sign up for yet. We are working through the verification model with a small number of teams running agents against real systems, and we would rather say that plainly than put a fake pricing page in front of you.
ARK Agent Readiness
Make your systems usable by agents.
Evaluate whether agents can safely and reliably interact with your applications, APIs, tools and workflows.
Where this is today: Available in two forms. The scorecard is free and takes a few minutes. The full assessment is a two-to-three week engagement run by ARC against your actual systems, and it uses the open scanners in ARK Labs as part of the evidence.
AI Fit
Determine where AI belongs before you build it.
Evaluate the problem, workload, economics, architecture and risks before committing engineering resources.
Where this is today: Available free on the web with no account and nothing stored. Deterministic rules match your description to a published catalogue of architecture shapes. A model never produces the recommendation or the score.
AI Fit Teams
Prepare the organization around the technology.
Understand AI readiness across teams, workflows, capabilities, governance and adoption.
Where this is today: Available as a three-to-four week assessment run by ARC. It is the organisational counterpart to AI Fit: the same scoring discipline, applied to the teams, workflows and habits that decide whether anything ships.
Flagship
ARK Control
Seeing a trace tells you what the agent attempted. Assurance asks a different question: did the intended outcome actually occur?
Where this is today: ARK Control is in design-partner development. It is not a product you can buy or sign up for yet. We are working through the verification model with a small number of teams running agents against real systems, and we would rather say that plainly than put a fake pricing page in front of you.
The operating contract
Agents need operating contracts, not just prompts.
An agent responsible for ongoing work needs more than instructions.
- Goal
- Trigger
- State
- Boundary
- Evidence
- Owner
A prompt is an instruction. A contract is something you can check afterwards and hand to someone else. The framework is free and published in full — there is no email gate on it.
- A defined objective
- What done looks like, in terms someone else could check.
- A trigger
- What starts the work, and what must be true before it does.
- Persistent state
- What it is allowed to remember between runs, and for how long.
- Enforced authority
- What it can reach, written as permissions rather than as guidance in a prompt.
- Evidence requirements
- What has to be produced to show the work was done.
- A human owner
- A named person, not a team inbox.
- Stop conditions
- What makes it halt rather than retry.
- Recovery behaviour
- What happens to work already in flight when it does.
Underneath
The layers that are not products yet.
Runtime, protocols, open tools and documentation. Each labelled with what it actually is, which in some cases is research.
Runtime
ResearchWhere a unit of AI work should execute, decided from capability, privacy, policy, latency and cost rather than from habit. Local, edge and cloud are a routing decision, not an ideology.
Protocols
Open sourceAgent-facing interfaces: MCP, agent-to-agent patterns, and the machine-to-machine conventions forming around them. We test against the published specs and publish what breaks.
Open tools
Open sourceScanners, catalogues and reference implementations from ARK Labs. MIT licensed, no account, most of them store nothing.
Documentation
Open sourceArchitecture notes, schemas and implementation guides, published alongside the code they describe.
Running agents you cannot yet verify?
Start with the free readiness scorecard, or book thirty minutes and we will tell you whether the problem is your interfaces, your authority model, or the fact that nothing confirms what changed.