For founders and CTOs shipping AI

Know where your
AI can fail.
Know what to fix first.

Nilsoft examines one agent workflow across tools, approvals and changing state, then gives your team an evidence-linked diagnosis and a prioritised remediation plan.

The Continuity Diagnostic. A clear next move for your engineering team.

ScopeOne bounded workflow
Fixed feeAUD $5,000
Delivery

Within five business days after required access, evidence and discovery are complete.

The demo is the easy part

Real workflows
don’t hold still.

The agent works. Then a person changes an order, a tool times out, or two systems disagree. Your team is left to work out what actually happened.

01 / CUSTOMER IMPACT

One request.
Two actions.

A retry can turn an uncertain outcome into a duplicate customer action. Who checks before the agent tries again?

02 / ENGINEERING TIME

Your team becomes
the reconciliation layer.

Engineers join logs, compare records and chase changing state by hand. The workflow only holds together while someone watches it.

03 / RELEASE CONFIDENCE

More capability.
The same hesitation.

You want to give the agent more responsibility. You cannot yet establish which decisions it should make or how to prove the result.

Inside a diagnosis / finding 01

An approval can be valid.
And already out of date.

Illustrative sample — synthetic data, not a customer result.

Refund agentEvidence excerpt
E01 / Approved state
order.version = 18
E02 / State at execution
order.version = 19
Precondition recheck
Not recorded

A changed fact. An unchanged permission.

The decision outlived its facts.

The agent may act on permission that no longer covers the current order. A valid earlier approval is not enough to establish that this action is still permitted.

Proposed correction

Bind approval to the relevant order state. Re-read and compare that state immediately before execution; route a mismatch back to the decision owner.

Inspect all three findings

What your team gets

Evidence to decide.
A plan to act.

The value is in the decisions the findings let you make: where the workflow breaks, which controls matter, and what engineering should do next.

  1. 01

    Understand what is happening.

    See the workflow as it runs, including its workarounds and material control gaps.

    Current-state workflow map · Continuity and control failure register
  2. 02

    Decide what must change.

    Identify who may act, what needs to be checked, and how the proposed workflow should behave.

    Authority / capability / effect map · Target governed transition model · Remediation architecture
  3. 03

    Give the team its next moves.

    Leave with three prioritised actions, their boundaries, and a way to verify each result.

    Three prioritised next moves · One 60-minute findings session
See everything in the diagnostic

Sean ManougeFounder · Systems Architect
Adelaide, South Australia

The person behind the work

A systems architect.
Accountable for the diagnosis.

Sean owns Nilsoft’s direction and system design. The work follows your workflow across tools, human decisions and system boundaries, then connects each finding to the evidence behind it.

Observe → resolve → act → prove. Where the evidence ends, uncertainty stays visible.

Meet Nilsoft

A bounded first engagement

Start with the workflow
you need to trust.

Before expanding autonomy, rewriting the wrong layer, or accepting more manual work: establish where the gaps are and what to fix first.

Read the scope and terms

Continuity Diagnostic / NIL-CD-001

$5,000 AUD · fixed fee

One bounded AI, automation, or operational workflow.

Within five business days after required access, evidence and discovery are complete.

Will you change production?
No. The diagnostic examines and proposes; implementation is separate.
Do we need to replace our tooling?
No. Existing evaluations, traces and logs can inform the diagnosis.
Are we committed to follow-on work?
No. Your team receives the findings and prioritised plan. Any follow-on engagement is optional.

Bring the workflow. Start the conversation.

What is your AI doing
that you can’t yet trust?

Tell us what it does, where you lose confidence, and why the outcome matters. We’ll establish fit and scope before any engagement.

Discuss your AI workflowOne workflow is enough to start.