AI development teamHow an app issue gets fixed.Click a step to see what happens. This is an example, not a live task.
Example issue · receipt photo upload
A receipt photo will not upload.
A tradesperson reports the problem in the app. It becomes a task for the dev team.
ReportedStep 1 of 7
Next: check the problem.
Gap reports reviewedReplacement builds under way; real-work trials still required
AI development team: testingPlanner is testing; Brain Atlas remains planned
NZ firstAustralia follows once ready
Current position
What is there. What still needs work.
Some parts are built but still need to be joined up and tested. This briefing summarises those gaps and the work to fix them. The pictures illustrate the workflows; current build status is listed below.
1 hour will not fit. Reprioritise—do not extend the working day silently.
Illustrative examples, not screenshots of the current builds. See the task statuses below for what is being tested.
See the proposed work
What this could look like in Keece.
Examples of the Simpro and Groundplan replacement workflows planned for Keece. These are proposed Keece screens, not working replacements.
KEECE■Office & Site / Job managementPROPOSED · SAMPLE DATA
Turn the enquiry into an agreed job.
Keep the customer, site, scope and price together. Bring checked drawing quantities into the quote, then carry the accepted work into the job.
Example jobOffice fit-out · Level 1
Customer and siteExample customer · Commercial office
Example stageQuote accepted
Agreed quote
Quoted work
Lighting, power and testing
Customer price
A$12,000
Estimated job costs
A$9,000
Accepted scope, drawings and quantities stay linked. Nothing needs to be typed into a second job system.
Proposed workflow only. All figures are made-up examples in A$, excluding GST. The same job history can support reporting, repeat service work and maintenance.
Build still required: join up the records, load credible prices and labour data, then test real orders and drawings. These examples do not change the replacement approval boundaries.
See the existing software references
Simpro job management
The Keece concept follows the full job: quote, crew bookings, site work, materials, invoices and costs. Purchasing is one part of that flow.
The Keece example follows the familiar pattern: drawing, count list, checking, then quantities and parts. The Keece concept includes a human review step for doubtful automatic counts.
Vendor material belongs to its respective owner. The Keece examples above are newly drawn concepts, not screenshots of a finished Keece feature.
01 / Delivery scope
Seven areas of work. Current position.
Planner is testing. The AI team, connections and replacement builds are progressing alongside it. Select an area for its status and remaining checks.
TestingKeece HQ · dev board, feedback and AI team
Set up the AI team and task board to build, test and track app changes.
Build work: testing
Keece HQ and the dev board are built for testing. Activate and prove the AI team’s end-to-end workflow with spending and approval controls.
Tasks and current position
□
Set up the dev task board
The dev board is now part of Keece HQ. Check the board, task ownership and review flow on real work. Global View belongs in Global BU.
Task leadBoard Chair
Testing
□
Set up the AI team
The Keece HQ build adds a factory master switch and pauses for temporary faults. Agent execution and spending remain gated while the team tests limits, approvals and one full task path.
Task leadPeople & AI Team Lead
Under way
□
Bring app feedback onto the board
App feedback feeds the HQ triage workflow. Test duplicate handling, prioritisation and the path from a report to a checked fix.
Task leadFeedback Lead
Testing
□
Set up dev team review
Review and approval flows are implemented. Prove them on a real change before relying on autonomous delivery.
Task leadRelease Review Lead
Testing
What we need
Servers that keep the AI tools running. AI accounts and test data. A spending limit and time for the dev team to review.
What must work before release
Take one real app issue all the way through. Fix it, test it, get dev team approval and put it live. Tell the person who raised it.
02 / Build order & time
Build it. Test it. Then use it.
Several work areas are progressing in parallel. These original estimates run from each area’s start, not from today.
Test Planner on real work while developing Keece HQ, AI connections and the replacements. Brain Atlas remains planned. Training follows each tested feature; this is not a strict one-after-another sequence.
01
AI development team
Testing
Build & agent tests
3–5 days
Trial on real work
2–3 days
Estimated total from start
5–8 days
Keece HQ and the dev board are built for testing. Activate and prove the AI team’s end-to-end workflow with spending and approval controls.
02
Brain Atlas
Planned
Build & agent tests
5–8 days
Trial on real work
2–3 days
Estimated total from start
7–11 days
Once the AI team is ready, list the systems and links. Then build and check the map.
03
Keece Planner
Testing
Build & agent tests
10–15 days
Trial on real work
5–10 days
Estimated total from start
15–25 days
Task imports, scheduling, team views and AI hand-off are built and testing. Site access is gated. Microsoft 365 sync and meeting bookings remain planned.
04
AI connections
Testing
Build & agent tests
5–8 days
Trial on real work
3–5 days
Estimated total from start
8–13 days
Connect the tools needed for Planner’s AI tasks first. Then move the remaining AI-page tools.
05
App training
Planned
Build & agent tests
5–8 days
Trial on real work
2–3 days
Estimated total from start
7–11 days
Set up training after the AI connections. Teach existing app features first, then add lessons as new features pass testing.
06
Simpro replacement
Under way
Build & agent tests
2–4 weeks
Trial on real work
2–4 weeks
Estimated total from start
4–8 weeks
First agreed NZ job functions only. Keep Simpro running during the trial.
07
Groundplan replacement
Under way
Build & agent tests
4–8 weeks
Trial on real work
2–4 weeks
Estimated total from start
6–12 weeks
Selected 2D counts, drawing edits, cable planning and line-level pricing are being tested. The full replacement remains under way. Keep Groundplan until estimator-checked results agree.
What the times mean
Build time includes coding, agent testing and fixing faults. Trial time is for checking the result on real jobs or drawings. The total adds these two periods together. A release follows only when the tests pass.
Working guesses, not deadlines
These are first estimates for the scope shown. They assume a working AI team, separate testing and prompt dev-team review. Agent capacity still needs to be measured; these ranges are not based on a fixed agent count. Re-estimate the remaining work from test results and the team’s improving coding efficiency. Training times cover the first lessons for all four roles; later features need more lessons.
Days mean working days; a week means five working days. Each range runs from that work area’s start, not from today. The task board assigns agents as the work needs them. Available compute, model usage, task dependencies and review capacity determine what can run at once. Normal setup is included; waiting for access, supplier data or review adds time. These are build-and-trial estimates, not dates for cancelling old software.
03 / Resources & cost
People, tools & approvals.
AI agents help build, test and record the changes. Keece staff still need to check the results on real work.
AI agents matched to the work
Keece HQ is being set up to coordinate agents as the work needs, within model usage, budget and review limits. Autonomous execution remains gated while it is tested. Claude Code and Codex are connections, not a two-agent limit. Keep changes in separate test copies and coordinate work on shared code.
Independent checks
Other agents plan the work and look into issues. Separate agents test the changes, put approved work live and update the training.
Dev team approval
The dev team reviews each test version. Changes to the AI task board need its approval before they join the main app. Any exception must be recorded on the task.
Working estimates: accounts and IT confirmation is pending. These figures are not an agreed budget or a spending approval.
App running costs
Starting estimates. Accounts will confirm.
These figures cover the app, not the whole development budget.
Earlier NZ baseline
A$400 / month
Hosting, data and AI for NZ work, including testing.
Group app running costs
A$2,850 / month
Approx. A$1,370 for tender AI, plus A$1,480 for automation, hosting, storage, database and email. Not an AI Gateway-only bill.
Full monthly budget
To confirm
Add dev and staff AI plans, existing software and support.
The A$400 and A$2,850 figures come from the 17 August 2026 briefing, not current invoices. The group estimate assumes roughly ten times the NZ workload. Both exclude GST, staff/dev AI plans, existing business software and staff time. A$2,850 a month is A$34,200 a year. Accounts will confirm the full bill. These historical estimates have not yet been recalculated for the newer models and more efficient workflows.
The wider return
Fewer bills. Less double-handling. Better checks.
The estimate above covers licence savings only. Keep cash saved, purchases avoided and time freed up separate.
Cash saved
Licence fees, outside take-off work or overtime we actually stop paying. Count only the amount removed, from the date it stops.
Purchases avoided
Software we would otherwise need to buy. CONQA was not purchased; that is an avoided cost, not money taken off today’s bill.
Time and quality
Less re-keying, faster drawing checks and more complete tender reviews. Prove the hours saved after checking and corrections. Time freed up is not automatically cash saved.
Existing software costs and possible savings
August briefing figures · A$ per year, excluding GST · not confirmed savings
Cost
What the briefing says
What needs checking
Simpro
A$97,000–155,000 across different licence scenarios. No invoice held.
Actual seats, discounts and bill. The lower scenarios already remove mobile licences; do not count that saving twice.
Groundplan
NZ account: A$2,412 a year, converted from NZ$242/month. The briefing reports an invoice; we have not checked that invoice here.
Other units’ actual bills. The A$20,000 group figure was a hypothetical purchase, not known spending.
Outside take-off work
About A$7/hour reported. Annual hours and spend unknown.
Hours, charges and checking time by unit. Compare against the full cost of checked AI output.
Invoices, active seats and cancellation dates. These are not a combined savings forecast or approval to replace more products.
Keep Simpro and Groundplan paid for while they are still needed. Licence savings start only when licences can be reduced or cancelled. No immediate full saving is assumed.
What accounts and IT need to confirm
Current software invoices by business unit, paid seats, currency, GST and contract end dates.
Staff Claude and other AI plans, seat counts and extra usage. Separate these from the dev team’s accounts.
Current app hosting, AI Gateway, storage, backups and support costs. Separate tests from normal work.
Outside take-off hours, checking and rework costs, plus representative completed drawings.
One-off build costs, staff review time, and the period when both old and new systems must be paid for. Hardware is not part of the current recommendation.
Compare the old and new costs for the same work and period. Show actual cash savings separately from time freed up. Keep unconfirmed costs out of any promised return.
Development AI subscription allowance
This earlier allowance is a reference, not a new subscription order. Review it against current pricing and measured usage as coding and API efficiency improve.
Historical subscription allowance
2–3 AI plans per developer
US$400–600 per developer / month
Working allowance: two or three Claude or ChatGPT plans per person, at US$200 per plan. This is not the full AI running budget. Tax and currency conversion are extra.
2 developers · 4–6 plans
US$800–1,200 / month
3 developers · 6–9 plans
US$1,200–1,800 / month
Recheck against actual usage: this allowance predates the latest efficiency improvements. Do not assume every developer needs every plan. Each developer needs their own accounts; this is not a shared-login plan.
What to check: usage limits, account rules and whether the task board can use the included allowance. API use—where software calls the model directly—is billed separately. Extra usage and app hosting are not included in this figure.
The dev team updates the task status and keeps a record of changes. Updates from the task board are not automatic yet.
Reviewed 25 September: recent merges add 2D drawing edits, improved counts, cable planning and line-level pricing to the Groundplan build. These tools are being checked on real drawings; the full Groundplan and Simpro replacements remain under way. Saved task statuses are updated separately and may lag code merges. Testing is not a rollout approval.
7 / 25
Build tasks in testing The earlier gap reports are not counted here. This tracks the build work listed above, not the whole existing app.
Testing: a build exists and is being checked; remaining tasks may still be planned. Under way: development or setup has started. Planned: work has not started.
25 matching tasks · last updated 2026-09-25
Loading saved information…
AI development teamSet up the dev task boardTask leadBoard ChairTesting+
The dev board is now part of Keece HQ. Check the board, task ownership and review flow on real work. Global View belongs in Global BU.
Task leadBoard Chair
AI development teamSet up the AI teamTask leadPeople & AI Team LeadUnder way+
The Keece HQ build adds a factory master switch and pauses for temporary faults. Agent execution and spending remain gated while the team tests limits, approvals and one full task path.
Task leadPeople & AI Team Lead
AI development teamBring app feedback onto the boardTask leadFeedback LeadTesting+
App feedback feeds the HQ triage workflow. Test duplicate handling, prioritisation and the path from a report to a checked fix.
Task leadFeedback Lead
AI development teamSet up dev team reviewTask leadRelease Review LeadTesting+
Review and approval flows are implemented. Prove them on a real change before relying on autonomous delivery.
Task leadRelease Review Lead
Brain AtlasList the systems and servicesTask leadKeece App EngineerPlanned+
Record apps, paid services, data, work processes, AI tools and key documents. Name the team responsible for each.
Task leadKeece App EngineerSupportProduct & Design Lead (brief)
Brain AtlasBuild the mapTask leadKeece App EngineerPlanned+
Create a 3D map with search, zoom and filters. Follow the links between systems. Include a simple list view.
Task leadKeece App EngineerSupportProduct & Design Lead (brief)
Brain AtlasKeep the map up to dateTask leadKeece App EngineerPlanned+
Explain each link. Show what has been checked and when. Clearly mark anything still assumed. Allow a linked text copy to be saved.
Task leadKeece App EngineerSupportProduct & Design Lead (brief)
Copy chosen Keece blocks into Microsoft 365. Let people book free meeting times. Add breaks, booking limits and options to change or cancel.
Task leadKeece App EngineerSupportProduct & Design Lead (brief)
Keece PlannerChoose who does each stepTask leadKeece App EngineerTesting+
Team delegation and AI hand-off are built. Check that results return to the task and a person approves the proposed completion.
Task leadKeece App EngineerSupportProduct & Design Lead (brief)
AI connectionsFind a new home for each AI-page toolTask leadTechnology LeadPlanned+
Keep the useful tools and records. Move them to existing Keece screens or make them available to Claude and ChatGPT.
Task leadTechnology LeadSupportProduct & Design Lead (screen design)
AI connectionsConnect the tools safelyTask leadPlatform EngineerTesting+
Authenticated Keece tool connections, Planner tools and Jev evaluation are implemented. Test each assistant’s permissions and approvals before wider use. Old-page retirement is separate.
Task leadPlatform EngineerSupportTechnology Lead (system design); Risk & Security Lead (access approval)
AI connectionsTest before removing the old pagesTask leadPlatform EngineerPlanned+
Keep screens for reviewing AI drafts. Make old links open the right new screen. Check the floating assistant still works.
Task leadPlatform EngineerSupportApp Test Engineer (testing); Risk & Security Lead (access approval)
App trainingCheck what training is missingTask leadProduct & Design LeadPlanned+
Check which app tasks are already covered. List what each role still needs.
Task leadProduct & Design Lead
App trainingBuild the lessonsTask leadProduct & Design LeadPlanned+
Reuse lessons where possible. Add screenshots with notes, short videos with captions and practice tasks.
Task leadProduct & Design LeadSupportTrade Language Lead (wording)
App trainingTrack who has finishedTask leadProduct & Design LeadPlanned+
Show lessons assigned, overdue and complete. Record which version each person finished. Update lessons when the app changes.
Task leadProduct & Design Lead
Simpro replacementCheck records, prices and ratesTask leadCommercial Systems EngineerUnder way+
The build uses Simpro history, labour norms and each business unit's rates to show quote cost, sell and margin with revisions. Check supplier prices, estimated cable rates and missing labour figures against real jobs. Purchasing, billing and cutover still need separate work.
Task leadCommercial Systems Engineer
Simpro replacementConnect purchasing, stock and billingTask leadCommercial Systems EngineerPlanned+
Join up orders, deliveries, bills, credits, stock and customer invoices.
Task leadCommercial Systems Engineer
Simpro replacementTest the first NZ tasksTask leadCommercial Systems EngineerPlanned+
Run timesheets, job costs and accounts through both systems. Check the figures match and that we can switch back.
Task leadCommercial Systems EngineerSupportApp Test Engineer (testing)
Simpro replacementKeep access to old recordsTask leadCommercial Systems EngineerPlanned+
Keep Simpro available for old records and any work still using it. Both systems may be needed for a year or longer.
Task leadCommercial Systems EngineerSupportData Engineer (data)
Groundplan replacementLink take-off to pricingTask leadTake-off & Estimating EngineerUnder way+
The build links drawing counts and measurements to quote lines. It now shows supplier prices and Jev estimates with their source, currency and confidence, and carries the estimator's supplier choice into pricing. Check each versus per-metre prices for linear lights before relying on totals.
Task leadTake-off & Estimating Engineer
Groundplan replacementCheck automatic symbol countsTask leadTake-off & Estimating EngineerUnder way+
2D reading now covers lights, outlets, switches where the legend is reliable, and boards and submains. The drawing page lets an estimator review confidence, add or move fittings and measure linear runs. Compare with checked drawings; unclear legends stay for manual review.
The build shows the board and circuit tree, lets people draw duct, conduit and tray, measures connected submains and plans shared-floor catenary for lights and power. Test routes and labour against real drawings. No undrawn duct or missing rate should be guessed.
Recent fixes avoid duplicate sheets and lights counted from architectural or security drawings. Quote revisions carry changed labour and prices. Re-run changed drawings and check quantities, supplier choices, price differences and approved materials before use.
Task leadTake-off & Estimating EngineerSupportApp Test Engineer (testing)
Progress notes
No progress notes from this build plan have been recorded yet.
05 / Accounts input
Put the real costs here.
The estimates above stay as they are. This separate table shows what changes when accounts checks the bills.
Monthly costs · A$ excluding GST · whole group
Today: current costs. Once in use: expected costs at today’s staff level. Unknown stays blank; enter 0 only when there is no cost.
How to complete the table
For per-user billing, enter a monthly rate and paid-user count. Expand a subscription with its + button and choose tradespeople, apprentices or other roles to create separate price rows. Parent totals then lock and add the rows up. Growth keeps each role’s current ratio unless you choose a manual forecast.
Pull from estimate copies a selected planning figure into a blank field, marked Estimate. It never replaces an entered amount or splits a total into invented role prices. Divide annual bills by 12; convert foreign bills to A$ and record the rate and date. Keep overlapping systems in the budget until their bills stop.
No invoices, staff names or account numbers here. Only added subscriptions and their role / tier rows can be removed.
Loading saved information…
Monthly saving: Complete the missing costs to calculate · example fields — saved figures are locked
This general briefing link cannot open company costs. Ask the dev team for the accounts edit link. Open it once, then enter costs here—changes save automatically. No app login is needed.
Staff and growth assumptions
Fills blank staff fields with 260 today and a 500-staff scenario; entered figures stay unchanged.
Start with about 260 staff. A forecast of 500 is a scenario, not a hiring plan. Per-user plans can keep today’s mix of roles and paid seats, or use a manual future seat count. The same person may use several subscriptions; do not add licence counts to get staff numbers.
Swipe sideways to compare prices. Use + beside a subscription to add roles or open details.
Monthly subscriptions · use + beside the name for roles, estimates and settings
Subscription / role
Billing
Today · A$ / month
Once in use · A$ / month
Figure status
Simpro
Groundplan
Trello
monday.com
Time4U
Training software
Other existing software
Keece hosting and database
AI used inside Keece
Staff Claude / ChatGPT subscriptions
Dev-team AI subscriptions and usage
Local AI machine running costs
Other ongoing support
Standard rows stay in place. Added subscriptions and plans can be removed. Do not also count an added subscription in Other existing software.
Extra one-off costs
Hardware purchase
Full installed cost, excluding GST. Enter 0 if this scenario does not buy a machine.
Build, setup and training
Extra paid work still needed. Exclude past spending and costs already in monthly bills.
Extra transition costs
Old-system overlap, migration and exit costs not included in the monthly comparison.
Live calculation · working estimates
Software and running-cost savings at 260 staff.
This calculator compares bills. It does not yet include the value of better site work, less admin or more jobs won.
Not enough figures for a return yet. Complete the missing prices, seats, names and one-off costs. Totals below add known amounts only; they are not a complete group cost.
Current monthly costs entered
$0.00
Replacement monthly costs entered
$0.00
Monthly saving at today’s staff level
Not ready
One-off costs entered
$0.00
First 12 months in use, less one-offs
Not ready
Time to recover one-off costs
Not ready
Return percentage needs complete figures and a one-off cost above zero. The first year means 12 months after the change is in use. Build time is extra.
Monthly saving = today’s bill − replacement bill. First-year result = 12 × monthly saving − one-off costs. Payback = one-off costs ÷ monthly saving. Staff time saved and purchases avoided are not counted as cash savings.
Separate growth scenario
What if Keece reaches 500 staff?
This compares both options at the same larger workforce. It is not extra savings to add to the return above. Per-user prices stay unchanged; plans marked for scaling keep today’s role mix. Fixed costs stay fixed. Usage forecasts need their own estimate.
Keep today’s systems — monthly known subtotal
$0.00
Use the replacement plan — monthly known subtotal
$0.00
Difference at the same workforce size
Not ready
Some forecast figures are missing. Open Growth forecast settings under each plan to check seats and usage.
Supplier price increases, minimum seats, volume discounts and extra hardware are not automatic. Enter them before treating this as a budget.