Evidence conflictDED-109
Which delivery record is right?
$400Both sources retained. The exact follow-up is ready.
JLJordan Lee · 3 days waiting
We already have agents. Is the work actually working?Reference workflow
Customer deductions
Customer deductions
AI WORK OPERATIONS
Know where the work is.
See what is progressing, what is waiting, where people are needed and what has actually completed.
Synthetic operating snapshot · SNAP-A25 Cases
13Ready for
Ready for
the next step
Inputs and current permission in place
13 ready6 need a person6 resolved
Where the work is waiting
6 need a personAuthority requiredDED-201
A credit needs the right authority.
$3,100Proposal prepared. An authorized approver reviews the exact request.
Proposed · No credit issued
Agents are capacity. The Case is the work.
Concept · No integrations connectedExisting enterprise agentField Runtime workerSpecialist modelDeterministic software
Work / Case
EvidenceAuthorityHistoryState
Verified outcome
Traditional observability: is software running?Field Runtime: is the work working?
Next: Attention 06 Cases need a personSynthetic operating snapshot · SNAP-A
HUMAN ATTENTION
Put human attention where it matters.
Humans enter for judgment, authority, missing evidence and genuine exceptions. Routine coordination stays with the work.
Evidence owner needed: resolve the source conflict or access gap after permitted retrieval has been attempted.
LONGEST WAIT · DED-109
Delivery proof conflict
Elapsed wait
3 days
Active effort
12 min
JLJordan Lee · Evidence owner
The smallest human contribution
✓
Permitted lookup
Delivery sources retrieved.
✓
Source comparison
Opposing claims retained. No precedence invented.
!
One targeted question
Ask the evidence owner for authoritative proof.
The next person gets prepared work, context and a specific question.
Waiting and working are different costs. Three days elapsed does not mean three days of human labor. This Case has consumed 12 minutes of active effort so far.
PERMISSION TRAVELS WITH THE CASE
Read evidence. Prepare the proposal. Hold at the boundary.
A confident worker cannot grant itself authority. New evidence and current permission govern the next step.
People contribute what only they can.The Case retains everything needed to continue.
Next: Outcomes Business proofSynthetic operating snapshot · SNAP-A
VERIFIED OUTCOMES
An agent saying “done” isn't an outcome.
Field Runtime separates what was reported from what was independently observed and what the business accepted.
AGENT / ACTIVITY METRIC
“Action completed”≠
BUSINESS OUTCOME
“Result independently observed and accepted.”Accepted: the business accepts the exact observed disposition. Other obligations and whole-case closure remain separate.
Later reliable contradiction → REOPENED
CASH COLLECTED
$18,000
Checked receipts · SNAP-A
VERIFIED*
Disputes resolved
6
Credits issued
$2,000
Open the proof
Sample synthetic Receipts*“Verified” is an illustrated work state, not a claim of live verification. These are sample Receipts, not the complete $18,000 ledger. Cash, credits and accepted dispositions can overlap; count each underlying cash effect once.
Outcome proof is the economic denominator.Count useful work that was observed and accepted.
Next: Economics Illustrative comparison · USDECON-A → ECON-B
25 identified Cases in each
25 identified Cases in each
ECONOMICS OF USEFUL WORK
Know what useful work costs.
The unit that matters: what did one verified, accepted resolution cost—including the work that remained unresolved?
Operating cost / accepted outcome
$37.50$25
Baseline → Later illustrative cohort
All-case cost ÷ accepted resolutions
Human attention / accepted outcome
30 → 15 min
50% less attention per accepted outcomeAccepted outcomes
20 → 24
20% more accepted resolutionsAI spend increased. Total cost of getting useful work done decreased.
All-case operating cost
$750 → $600Human effort · all Cases10 hours → 6 hours
AI + runtime cost$150 → $240
Human cost$600 → $360
Median accepted cycle time
3.0 → 1.8 days
40% shorter in the illustrative comparison.
Unresolved Case ages remain a separate measure.
CAPACITY RELEASED
4
person-hours
Released for higher-value work
The illustrated reduction in active effort creates capacity for other work. Payroll savings are not established.
Illustrative comparison. Setup, support and fees are unknown. Cohort comparability and causality are not established. These cohorts are separate from SNAP-A and the CHANGE-A improvement proposal.
Inspect the metric basis & complete comparison
| Metric | Baseline · ECON-A | Later · ECON-B |
|---|---|---|
| Identified Cases | 25 | 25 |
| Accepted resolutions | 20 | 24 |
| All-case human effort | 10 hours | 6 hours |
| Human attention / accepted | 600 min ÷ 20 = 30 min | 360 min ÷ 24 = 15 min |
| Human cost · $60/hour | $600 | $360 |
| AI + runtime | $150 | $240 |
| All-case operating cost | $750 | $600 |
| Operating cost / accepted | $750 ÷ 20 = $37.50 | $600 ÷ 24 = $25 |
| Median accepted cycle time | 3.0 days | 1.8 days |
| All-in delivery cost | Unknown | Unknown |
Every Case contributes effort and cost—including review, rework, failed and unresolved work. Only accepted outcomes enter the denominator. This prevents unsuccessful work disappearing from unit economics.
The human-cost rate is illustrative. AI + runtime is a combined category; model/tool versus infrastructure cost coverage still needs validation. A comparable window, Case mix and outcome definition would be needed before attributing benefit.
Different metric: SNAP-A’s $450 accrued ÷ 25 identified Cases = $18 per identified Case at a point in time. It is not the $25 cost per accepted resolution above.
Inference cost is not the business denominator.
What did one verified unit of useful work cost?
The outcome definition belongs to the workflow.
Conceptual future units · No performance claims
Accepted resolutionVerified claimCompleted procurement requestResolved escalationApproved renewalDeployed change
Spend follows verified work.Now decide what to improve and where to invest.
Next: Improve Candidate method · Future capacity decisionCHANGE-A
Production effect: not measured
Production effect: not measured
IMPROVE & ALLOCATE
Turn observation into a better operating method.
Use workflow evidence to propose a change, evaluate it, obtain approval and measure the effect before allocating more.
Observe →Propose →Evaluate / approve →Measure →Allocate
Earlier delivery-proof retrieval
Proposed · Not releasedCURRENT METHOD
Evidence arrives late.
01Review the invoice
02Pause and chase missing proof
03Cash and contact plans wait
PROPOSED METHOD
Check permitted proof earlier.
01Check access and source limits
02Retrieve and compare the proof
03Continue or prepare a targeted ask
Inspect illustrative historical Replay24 passes1 overall verdict pending
25 synthetic historical Cases · No replay executed hereExternal actions disabled
DED-101 · Earlier proof requestPass
DED-102 · Source retainedPass
DED-109 · Safe-hold assertionPass
The hold worked. The Case stays held.
DED-109’s overall verdict is still pending. A passed assertion does not complete the remaining evaluation.
DED-109’s overall verdict is still pending. A passed assertion does not complete the remaining evaluation.
Owner approval is pending. The shadow test has not started. Method approval would not refresh old Case consent or grant financial authority.
Production impact: not measured. Establish evidence-owner effort, duplicate requests, payment waiting and contact conflicts. Financial authority stays unchanged. ECON-A/B does not prove the effect of this proposal.
ILLUSTRATIVE FUTURE CAPACITY DECISION
Where should the next dollar of intelligence—and the next human hour—go?
HUMAN-ATTENTION CAPACITY LEFT
8 person-hours
Team-declared planning availability
PROPOSED FUNDING
$7,500
Not committedBEFORE COMMITTING · CUSTOMER DEDUCTIONS
Cost a smaller experiment first.
Earlier permitted delivery-proof retrieval. Confirm access, baseline and the cost of getting to an accepted resolution.
Candidate experiment: Replay 5 completed Cases from this workflow. Establish a baseline; do not commit funding from these illustrations.
Work waiting$30,000 disputed · Count unknown
Desired useful outcomeAccepted dispute resolution
AI/runtime cost of experimentNot estimated
Evidence before allocating moreAccess, proof, owner effort, full cost
Select a workflow to inspect the decision basis. Funding figures are planning proposals, not AI-spend estimates. This 8-hour plan is separate from the 4 person-hours released in ECON-A/B. No allocation engine is operating in this preview.
One operating loop. The work stays connected.
- 01AI Capacity
- 02Work Runtime
- 03Governance &
Human Attention - 04Verified Outcomes
- 05Operational
Observability - 06Economics
- 07Evaluation &
Improvement - 08Capacity Allocation
↺ Allocation informs the next use of capacity. Owners retain authority; changes require evaluation and approval. Future product concept.
See this on your work.
Start with 5 completed Cases from one recurring workflow. See what AI could carry, where people still matter, what a verified outcome means and which operating metrics are worth establishing.
What useful work can we afford next?Begin with one workflow and a small, reviewable experiment.
Back to Operations