AI your governance committee doesn't have to fear
Large organisations don't have an AI problem, they have an AI-approval problem. JiraVision was built for that gate: read-only until a named human approves, every metric opens to show its formula, and a compliance story told in the present tense — including the parts that are 'in progress'.
- No autonomous writes — approvals are logged and attributable
- Wellbeing and load signals aggregated before anyone sees them
- SOC 2 and GDPR alignment in progress; ask us specifics, get straight answers
The problem
Enterprise delivery tools get blocked three ways
Not by features — by trust. The AI that touches production systems, the metric that ends up in a compensation review, the vendor whose compliance page says more than it can prove.
Autonomy is the non-starter
An AI that changes sprint scope or reassigns work unilaterally fails security review on principle, before anyone reads a feature list. Correctly.
Metrics get weaponised downstream
Per-person numbers leak from dashboards into performance conversations nobody authorised. Once that happens, the team games the metric and the data dies.
Compliance theatre wastes everyone's time
Vendors claim badges they don't hold; procurement burns a week discovering it. We'd rather hand you an honest 'in progress, here's what we can show you today'.
What you can vet
Controls, not promises
Every claim on this page maps to a behaviour you can test during a pilot. If it isn't shipped, it's labelled planned.
Approval-gated writes
JiraVision drafts; nothing reaches your Atlassian tools until a named approver acts — and the approval is recorded with who and when.
Read-only connections
Today the live integration is Jira, read-only. GitHub, GitLab and Slack are roadmap items and are labelled as planned, not available.
Team-level aggregation
Wellbeing check-ins merge into team figures before display; small groups are combined, never shown raw. Individuals see their own data.
Formula-on-card metrics
Predictability, load and fairness numbers expand to numerator, denominator, source events and exclusions — built for the question 'how do we know?'
Data stays yours
Workspace data stays in your database, is never used to train public models, and can be exported or deleted on your schedule.
Straight answers on compliance
SOC 2 and GDPR alignment are in progress for an early-stage team, not certificates we hold. Bring specifics to /contact and get a yes, a no, or a date.
The stack
Four products, one governance model
AI Scrum Master
Capacity-aware planning drafts with per-item rationale — a proposal workflow, never an autopilot.
Ethical Metrics
A deliberately small metric set with published formulas and a refusal list of what we won't measure.
Team Wellbeing
Early-warning load signals that stop at the team boundary by design.
Gamification
Opt-in recognition with guardrails that make the surveillance trade-off visible before you can turn it on.
01
Human approval is the only write path
Treat this as an architectural property you can test, not a policy promise. Every mutation — sprint content, reassignment, report publication — terminates in a human approval in your workspace. The draft itself is reversible in one action, and the log names the approver. When your auditors ask 'what did the AI do?', the answer is 'nothing, on its own — here is who did, and why'.
- Read-only until a named approver acts
- Approval log attributable to people, not tooling
- One-click revert of a whole draft, including mid-sprint
02
Metrics that survive an audit
Delivery numbers inside big companies face two audiences: the team who must trust them and the review that must defend them. Every metric opens to its formula, source events, exclusions and last recalculation. That makes disputes editable — you argue about a definition, log the change, and everyone sees it — instead of political.
- Numerator, denominator and caveats printed on the card
- Definition changes are logged and visible to the org
- A published refusal list: no lines-of-code, no per-person velocity
03
The compliance conversation, held honestly
We are an early-stage team. SOC 2 and GDPR alignment are underway — that is the present tense, and we won't sell you the future tense. What we can show you today: the data-flow diagram behind this page's method section, the approval logs, the aggregation model, and a working session where your security questions get a direct answer or a real date.
- Certification status stated plainly, in progress and named
- Pilot scoped to one value stream while paperwork catches up — honestly
- Sub-processors, retention and DPA specifics via /contact
Sprint planning sandbox
Watch a sprint overload itself — then fix it, one approved move at a time
Auto-draft fills the sprint optimising for skill fit and nothing else. Someone quietly ends up carrying the plan while someone else waits. Propose a rebalance, read every move, and apply only what you agree with.
Runs entirely in your browser on sample data — in the product this reads your real Jira capacity.
Backlog · unassigned
Nine sample tickets, one sprint, no crystal ball.
- Auth token refresh flow5 ptsJV-101backend
- Sprint retro summary view3 ptsJV-104frontend
- Export CSV endpoint2 ptsJV-108backend
- CI pipeline speedup3 ptsJV-112infra
- Onboarding empty states4 ptsJV-115frontend
- Regression pack for billing3 ptsJV-118qa
- Search ranking fix5 ptsJV-121backend
- Staging environment rebuild5 ptsJV-124infra
- Wellbeing check-in modal3 ptsJV-127frontend
Team load spread
100100 minus how unevenly points land across people, with a penalty for anyone over capacity. A low score means someone's week is quietly already lost.
Ada Kwarteng
frontend · qa · 8 pts capacity
- No work assigned — capacity left on the table.
Jonas Meier
backend · 14 pts capacity
- No work assigned — capacity left on the table.
Tessa Pham
infra · 8 pts capacity
- No work assigned — capacity left on the table.
Rin Sato
frontend · 12 pts capacity
- No work assigned — capacity left on the table.
Deterministic draft: best skill match first, then smaller tickets, then backlog order — no randomness, same input, same plan.
How it works
A procurement-shaped pilot, not a science fair
Most enterprise deployments fail in the gap between the demo and the review. Start from the review.
- Step 1
Bring the hard questions early
Start at /contact with your security and data-flow questions. You get answers in the present tense, including what isn't available yet.
- Step 2
Scope one value stream
Pick one team or programme. Read-only Jira access, one approval workflow, a defined exit. Nothing else enters production.
- Step 3
Run the controls check
Test what this page promises: approval logs, aggregation behaviour, formula transparency, export and delete. Bring your own auditors.
- Step 4
Decide on evidence
Expand per division or walk away with the export. Predictability and load trends from the pilot are yours to keep either way.
Straight talk
Be honest about the fit
Works well if
- Organisations on the Atlassian stack wanting forward planning and fairness signals, not another reporting layer
- Buyers who prefer an early-stage vendor's straight answers over a slide of borrowed logos
- Governance models that require human-in-the-loop AI as a hard constraint
Not for teams that
- Anyone who needs a SOC 2 report or signed DPA this quarter — alignment is in progress; we won't pretend otherwise
- Programmes requiring on-premise deployment today — ask us via /contact before assuming
- Teams whose actual requirement is HR case management or therapy services — this is workload instrumentation, not care
Who it's for
Who has to say yes
Enterprise adoption is three signatures. This product was designed at each of their desks.
Security & risk
A vendor whose AI cannot write without approval, whose integrations are read-only, and whose compliance gaps are admitted before you find them.
- Approval-gated architecture
- Data never trains public models
- Export/delete on your schedule
Procurement
Claims you can verify in a pilot, roadmap items labelled planned instead of implied, and per-seat terms listed on /pricing.
- Testable claims
- Honest status language
- No mystery line items
Engineering leadership
Portfolio predictability and load balance with methodology one click away — and signals that stop at the team boundary so metrics don't poison trust.
- Defensible numbers
- Team-level by default
- Drift alerts before incidents
No black boxes
What your security review will read
The relevant facts, in one place, present tense.
- Access model
- Jira is the live integration and it is read-only. Writes occur only through an approval executed in your workspace, and each approval is logged with approver and timestamp.
- Data handling
- Workspace data stays in your database. It is never used to train public models. The AI layer is IBM Granite, chosen for enterprise data handling; every suggestion links to the reasoning and source data behind it.
- Privacy model
- Wellbeing and load signals aggregate to team level before display; groups below the minimum size merge. Individuals always see their own detail; nobody else's manager does.
- Compliance status
- SOC 2 and GDPR alignment in progress — not certifications held today. Full status, specifics and dates: /contact. Planned features across the site are labelled planned.
Included in every plan
This product ships with the platform — Starter, Professional and Enterprise all include it. Pricing scales by seat, annual billing saves 20%.
What early teams say
“The approval log is what got us past the architecture review. 'The AI did it' is not a sentence our audit trail can contain — here it structurally can't happen.”
“They told us SOC 2 wasn't done, told us when it would be, and let our pilot proceed with those limits written down. That honesty is why we're still talking.”
Example quotes from early-access conversations. We publish attributed testimonials only with permission.
Shipped
Shipped against this page's claims
2026-10-23
Feature
AI sprint reports
Capacity-aware sprint drafts now ship with a written rationale per suggested item, so planning decisions can be reviewed and edited before approval.
2026-09-30
Improvement
Wellbeing signals move to team-level aggregation
Check-in and workload indicators are now aggregated before display by default. Individual responses are never surfaced raw.
2026-08-28
Feature
Ethical metrics methodology view
Every fairness and workload metric now links to the exact formula and data window behind it.
Keep looking
Solutions for other teams
Bring your hardest questions first
Start the working session at /contact: security model, compliance status in the present tense, and a single value-stream pilot you can evaluate on evidence.