Skip to content
How we work

Audit. Pilot. Operate.

A staged model with a gate at every step, so you only invest further when the evidence says it's worth it. No long contracts up front, no black boxes at the end.

The engagement model

Four stages, three gates.

  1. 00

    Free workflow audit call

    We look at one recurring workflow together and tell you honestly if it is worth automating.

    Gate: Is there a fit?
  2. 01

    Audit

    Fixed scope. Process maps, baseline metrics, data-flow map, prioritised opportunities and architecture.

    Gate: Business case agreed?
  3. 02

    Pilot

    One workflow, built at low autonomy, measured against the baseline with exit criteria agreed upfront.

    Gate: Success criteria met?
  4. 03

    Managed retainer

    We operate, monitor and improve, with monthly reporting on hours returned, quality and cost per task.

Stage 00 · You + a Mirror.lk lead

Free workflow audit call

We look at one recurring workflow: what it costs in time, where it breaks, and whether automation or an agent would help. You leave with an honest view either way — including when the answer is “not worth it yet”.

You receive
  • 01Short written summary of the opportunity
  • 02Recommendation: proceed, wait, or don't
Stage 01 · Fixed scope, agreed in writing

Audit

We map the workflow end to end, measure the baseline, classify the data involved and design the target system — including which steps stay human.

You receive
  • 01Process map and baseline metrics
  • 02Data-flow map and list of processors
  • 03Prioritised opportunities with value and risk
  • 04Architecture recommendation and pilot plan with exit criteria
Stage 02 · One workflow, measured

Pilot

We build one workflow at low autonomy (draft or act-with-approval), test it against an evaluation set, and run it alongside your team against the agreed baseline.

You receive
  • 01Working pilot in your systems
  • 02Evaluation results
  • 03Approval flow and exceptions queue
  • 04Cost-per-task dashboard
  • 05Scale-or-stop recommendation
Stage 03 · Only if the pilot meets its criteria

Managed retainer

We operate, monitor and improve the system — raising autonomy where accuracy is proven, adding workflows as the results justify them.

You receive
  • 01Monthly report: hours returned, quality, cost per task
  • 02Ongoing QA and incident handling
  • 03Quarterly roadmap review
Autonomy levels

How much an agent is allowed to do — and who checks.

Every agent is assigned a level before launch. Controls scale with the level, following the proportional governance approach Gartner recommends.

  1. L0AssistHuman: Does the work
  2. L1DraftHuman: Edits & sends
  3. L2Act with approvalHuman: Approves each action
  4. L3Act within guardrailsHuman: Reviews exceptions & samples
  5. L4OrchestrateHuman: Owns outcomes & policy
Recommended for most small-business workflows · autonomy rises only on measured accuracy
Source: Gartner, 26 May 2026(opens in a new tab)
Measuring results

Hours Returned to Humans.

The outcome we report every month. Defined before the pilot, measured from logs, net of the time spent supervising the system.

MethodHours returned = (tasks completed or assisted × baseline minutes per task ÷ 60) − oversight hours
  • —Baseline minutes per task and monthly volume are measured over two weeks before the build.
  • —Completed and assisted tasks come from agent and automation logs, not estimates.
  • —Oversight hours include reviewing, approving, correcting and handling escalations.
  • —Quality is reported alongside: escalation rate, correction rate and QA findings.
Why cost discipline matters

Cheaper tokens, bigger bills.

Two 2026 Gartner forecasts shape how we design every engagement.

>5×

forecast rise in inference cost per agentic workflow through 2028

70%

of enterprises forecast to abandon agentic AI built by vendor forward-deployed engineering by 2028

Principles

What you can hold us to.

You own the build

Code, workflows, prompts, evaluation sets and documentation are handed over, and accounts sit in your name wherever the platform allows.

Autonomy is earned

Agents start supervised. Independence follows evidence from evaluation sets and production QA — and can be reduced if quality drops.

One human accountable

Every workflow has a named Mirror.lk owner and a named owner on your side.

Cost you can see

Per-task budgets, alerts and a monthly cost-per-completed-task figure. No surprise usage bills.

Written down

Decisions, runbooks and escalation rules are documented so anyone on your team can understand the system.

Honest scope

If a workflow isn't ready for automation, we tell you — and what would make it ready.
Next step

See where AI can return hours to your team.

Book a free workflow audit call. We'll look at one recurring workflow with you and tell you honestly whether automation or an agent is worth it.