We redesign the workflows your company runs on.
Umbra Studio is the consulting arm of Umbra Group. We work with operators on high-friction, compliance-heavy, judgment-dense workflows — the ones AI hasn't touched because the stakes are too high. Humans stay above the loop. Models stay on task. Governance stays non-negotiable.
The technology can increasingly be found everywhere — except on the bottom line.
McKinsey Growth, Marketing & Sales Practice, Reinventing marketing workflows with agentic AI, April 2026.
Series compiled in Studio's internal McKinsey Canon. The phrase “Lighthouse” originates with Krivkovich, The McKinsey Podcast, April 2026.
Three shapes of engagement. One discipline.
- Phase 01 · Observation (weeks 1–2)
- Phase 02 · Redesign (weeks 3–5)
- Phase 03 · Build & ship (weeks 5–9)
- Phase 04 · Handoff (weeks 9–10)
- Foundation maintenance · eval re-runs · drift
- Governance Runbook revisions
- The +1 improvement · one per month
- Monthly health report · Slack channel · quarterly review
- Day 01 · Principles & pattern library
- Day 02 · Candidate mapping & scoping
- Roadmap & risk register
- 30-day async follow-up
Four phases. Ten weeks. One working system. Then continuous.
The Sprint is a Forward Deployed Engineering engagement: the engineer team embeds inside your workflow for ten weeks and ships a working system into your infrastructure.
Observation. We don't touch anything yet.
Shadow the actual humans doing the actual work. Read the emails, sit in the reviews, watch the exceptions, log where the week disappears. The goal: a map of the workflow as it actually operates — not how it's documented.
Deliverables: workflow map, friction log (ranked by hours lost per week), risk register, candidate-agent scope with non-negotiables called out.
Redesign. The loop humans sit above.
Architect the new workflow with the 8 Lighthouse principles as forcing functions. Every step: what the agent does, what the human approves, what the audit trail captures, what triggers a handback.
Deliverables: target-state architecture, agent specs, prompt + tool library, eval suite, governance gates, rollout plan.
Build & ship. Production from day one.
Build the real system against the real workflow. Agents ship behind gates. Humans approve via the interfaces we build. Every run logged, every decision auditable, every exception handed back cleanly.
Deliverables: production code, eval dashboards, approval UI, runbooks, incident playbooks. No throwaway demos.
Handoff. You own the instrument.
Full handoff to your team. Training sessions, runbook walkthroughs, governance review, architecture decision record. Default next step: transition into Lighthouse Watch — the productized monthly service that keeps the system stable, current, and improving.
Deliverables: runbook set, ADR, operator training (3 sessions), 30-day warranty window, Watch transition.
Lighthouse Watch. The system that keeps getting better.
The Sprint shipped your system. Watch keeps it stable, current, and improving — month after month, while the foundation layer underneath churns. New models drop, evals drift, governance ages, library patterns sharpen. Watch is the channel through which all of those reach your system without you commissioning a new project.
Every month: model swap testing, eval re-runs, governance updates, drift detection — plus one delivered improvement (the +1), eval-gated, drawn from the Studio pattern library or a measured workflow optimization. Plus a 3–4 page health report on the 1st of every month, signed by the lead engineer.
Seven platforms. One cockpit.
Every Sprint deploys a subset of these. They're not pitches or roadmap items — they're operating in production today on a live reference deployment. One cockpit (Atlas) sits above six executors that do the work.
Flagship · Cockpit · Meta-agent
Atlas
A meta-agent with memory that watches the stack, briefs you each morning, learns from your feedback, and routes work back to you when it needs judgment.
Open Atlas spec →Five surfaces
- Morning BriefA.1
- ChatA.2
- KnowledgeA.3
- FeedbackA.4
- WatchersA.5
SchemaForge
Makes every page legible to search engines and AI assistants.
On-page SEOE.2OnPage Doctor
On-page health diagnosed and repaired in continuous cycles.
Entity hubsE.3HubForge
Entity hub pages auto-created and maintained with authority signals.
DistributionE.4PressRelay
Multi-channel distribution with per-channel recovery built in.
DiscoveryE.5PulseRadar
Market scan, ranked surfacing, drafted response.
Weekly recapE.6Loopback
Closes the loop on every week of content. Performance compounds.
We tested the framework on our own newsroom first.
Indietheka is the longest-running Spanish-language indie music outlet still in operation, and operationally the longest-running Forward Deployed Engineering engagement Studio runs in the LATAM mid-market — founded June 2012, run by one editor with no outside capital. The Lighthouse Sprint in Q4 2025 produced 16 agents. By Q1 2026 the fleet had moved off Cowork sessions onto durable infrastructure: 26 workflows running 24/7, a Slack feedback channel for post-publish corrections, and an operational dashboard. In Q2 2026 we built Atlas on top — a studio-level orchestrator that reads the fleet, classifies editorial feedback, commits corrections back to the repo, and dispatches work to Cowork when a human touch is needed. Eight reusable patterns and integrations — durable publisher, multi-channel fanout, scheduled audit with HITL, cron discover-rank-publish, plus the Spotify, Slack and GitHub modules — have been externalized as @umbra/* packages.
▸ Measured outcomes · 16-agent production system
▸ Atlas · the layer above the fleet
"We didn't automate the editing. We automated everything that wasn't editing. Then the editing got better, because the editors had time to edit."
Short answers to the questions we get.
We build. The Sprint ships a working system in production against your real workflow. The Workshop is advisory; the Sprint and Watch are engineering work. Staffing is scoped to the workflow — every engagement carries dedicated engineering, design, and governance ownership for the Sprint duration.
Three things. One, we run on the Lighthouse framework — a public, opinionated discipline, not a bespoke methodology per client. Two, we built it on ourselves first (Indietheka) before selling it. Three, we ship to production. The deliverable is code against the real workflow, not a strategy deck.
If you want a chat UI on top of your knowledge base, we're not a fit. If you want someone to manage an offshore dev team, we're not a fit. If you want a 30-page strategy deliverable without code, we're not a fit. We redesign workflows and ship systems. That's it.
Yes — the Beacon Workshop (2 days, $18k) is designed for that. You leave with a ranked roadmap, risk register, and clear scope for a Sprint if we decide to continue. About 40% of Workshop clients move to a Sprint within 90 days.
Sprints range $60–120k depending on workflow complexity, integration surface, and sensitivity (compliance, PII, financial). Watch is tiered: $14k/mo Essentials, $24k/mo Standard (default), $38k/mo Pro. The Workshop is $18k flat. Sprint+Watch bundle: $290k flat for the Sprint plus the first 12 months of Watch · Standard. No hourly billing.
We take two concurrent Sprints, staggered. As of Q2 2026 we have one slot open for Q3 and one for Q4. Watch and Workshop calendars are less constrained — we can typically start a Watch engagement within 2 weeks of signature.
Two Sprint slots open for Q3–Q4 2026.
Tell us about the workflow that eats your week.
Most engagements start with a 30-minute scoping call. Send over your context — what the workflow is, who runs it, what's broken, what isn't — and we'll reply within 48 hours with availability, a rough read, and next steps.