Hardware & Workspace.

Hardware is where most studios either over-spend on theatre or under-spend on reliability, and both failures look the same in an engagement retrospective — a demo that didn’t ship, a client call that dropped, a migration that corrupted because a laptop failed without a backup. The goal of this doc is to get hardware decisions off the founder’s plate and onto a set of named tiers with named refresh windows, so the next decision is “time to refresh the Sprint Lead’s machine” rather than “what laptop should we buy this year?”

Our baseline philosophy: buy quiet, reliable, portable, Apple silicon. Every Studio role runs on Mac. Every role has a named tier. Every tier has a named refresh window. Everything else in this doc flows from that three-line policy.

L1 Physical decomposes into seven things we own: workstations, displays, peripherals, mobile QA fleet, network and power, backup infrastructure, and the travel kit. Each has a named tier policy and a refresh window. Everything is capex, tagged in Wave, and rolls up to the Year-N hardware envelope below.

Three tiers, named by the workload they serve. Every Studio role maps to exactly one tier. Tier is a function of what the role does day-to-day, not of seniority — a senior Sprint Lead and a junior Sprint Lead are both tier B, because the workload is the same shape.

Tier A is the only tier that prioritizes raw compute. The 64 GB unified memory is the line item most Studios get wrong by under-speccing — once you run Docker, a local model for offline testing, and a full IDE+browser workload concurrently, 32 GB becomes a constant swap battle. Over-spec memory, under-spec storage (we lean on iCloud + the backup stack in §07 for spillover).

Alternatives. (1) Mac Studio M5 Max + 16” MacBook Air as companion — viable if the Agent Engineer is > 80% desk-bound; adds ~$1,800 but gives headroom for heavier local inference. (2) 14” M5 Max instead of 16” — acceptable for an engineer who prefers portability; accept the slightly worse thermals. Not acceptable: M5 Pro instead of Max (memory ceiling too low), or any PC workstation (breaks our single-OS discipline).

Tier B is the Studio’s most-bought tier. Balanced for a role that spends the morning in an IDE and the afternoon on a client call from a client site. 36 GB is the floor — not 18, not 24. The difference between 18 and 36 shows up the first time a Sprint Lead opens Chrome with 40 tabs, Figma, VS Code, Zoom, and a Docker container concurrently.

Alternatives. (1) Base M5 Pro with 24 GB — $600 cheaper, accept that memory pressure will show up in month 6 and refresh earlier. (2) 16” M5 Pro — acceptable if the Sprint Lead prefers a larger daily-carry; adds ~$400 and ~0.6 kg. Not acceptable: MacBook Air in this role; the sustained workload thermally-throttles the fanless design.

Tier C is for roles that never run Docker, rarely open an IDE, and spend most of their day in a browser, docs, and Zoom. The Air’s fanless design is an asset here — the Studio Lead is on more client calls than anyone else in the Studio, and the audible fan on a Pro is a real and measurable distraction.

Alternatives. (1) Mac mini M5 + external display — viable for a pure desk role (a permanent QA station or a demo rig) and the cheapest option in the lineup at ~$1,100. (2) 13” MacBook Air — acceptable for travel-heavy roles but the smaller display meaningfully hurts finance/spreadsheet workflows.

The desk setup is one standard for all tiers. The laptop differs; everything else is identical. This makes onboarding trivial (“you know this desk, you’ve used this desk”), and it means any workstation can be swapped into any desk during a failure or refresh without reconfiguration friction.

Every desk meets the same ergonomic standard: display at eye-height, wrists neutral, chair adjusts in five axes, feet flat to the floor. The Studio default chair is a Herman Miller Aeron (size B) secondhand or an IKEA Markus on the budget tier. Desk height: electric sit-stand, 64–128 cm travel range, programmed to two positions per user.

Most of what the Studio builds is back-of-house agent infrastructure, not mobile apps — but every engagement has at least one surface a human uses from a phone: a Slack alert, a Lnk.Bio page, a short Loom, a preview URL the Executive Sponsor opens on their commute. A minimal shared QA fleet catches the class of bug that only shows up at 375 pixels wide.

A cloud device farm (BrowserStack, Sauce Labs, LambdaTest) is an acceptable alternative to a physical fleet when the Studio is < 2 people. Monthly cost ~$40; trades the capex of ~$2,800 for a recurring opex line. Why we still run physical devices: the haptic and network realism of a real device catches bugs a cloud emulator misses — specifically anything involving PWA installs, push notifications, or real cellular network throttling. Past year-two headcount, run both.

Network and power are what clients never see until they break, and they are the single most common cause of “the Friday demo failed” retrospectives in the industry. The Studio runs a belt-and-suspenders network — primary fiber, LTE failover, UPS on every critical desk — because an hour of downtime during a Gate call is worth more than a year of the monthly cost.

We run a UniFi stack: a Dream Machine Pro as the gateway, one or two U7 Pro access points depending on desk count. VLANs split Studio workstations from the QA fleet from IoT (printers, cameras). SSIDs: umbra-studio (primary), umbra-iot (QA fleet + printers), umbra-guest (clients, visitors; rate-limited, no LAN access). Per-client VPN allowlisting uses the Dream Machine’s static IP routing.

Three copies of important data, on two media, with one off-site — the classic 3-2-1 rule, applied at the workstation level and the Studio level. Without it, a single laptop theft becomes an existential incident. With it, the worst case is a two-hour restoration and a quarterly post-mortem entry.

A Synology DS224+ with two 8 TB WD Red drives in SHR (essentially RAID 1) is the Studio’s local second-media target. Lives on the UniFi rack. SMB shares: timemachine/, archive/ (the frozen engagement archive per §06 of USO-ST-04), vault/ (contracts mirror). Drives are rotated on a 5-year schedule; we replace one at a time and let SHR resilver, so the NAS is never offline.

The travel kit exists so that a Sprint Lead who flies to a client site on Monday morning never has to improvise. The same bag, the same cables, every trip. Weight target: < 4.5 kg including laptop. Total kit cost: ~$700 excluding the laptop. Refreshed every 36 months or sooner if a strap fails.

The purchase is the easy part. What most studios skip — and what causes year-three problems — is the asset register, the refresh trigger, and the disposal path. This section defines all three.

Every Studio device — Mac, iPhone, iPad — is enrolled in Jamf Now (free tier up to 3 devices, then ~$4/device/month). MDM handles: remote wipe, FileVault recovery-key escrow, app deployment, OS update enforcement. The alternative, Mosyle, is broader (Apple-only and a touch cheaper) and an acceptable swap; we picked Jamf for distributor familiarity.

One Airtable base, one row per device. Fields: asset_id, model, serial, owner, purchase_date, applecare_end, assigned_desk, refresh_due, status (active / staging / disposed). The refresh_due column auto-calculates 30 months after purchase for Tier A/B, 36 months for Tier C. The Studio Lead’s quarterly checklist includes a glance at any row where refresh_due is within 90 days.

The Studio Lead’s recurring hardware work collapses into the list below. If this list runs on schedule, the rest of this doc stays invisible — which is the point.