Company operating system strategy
Yes: building every company tool ourselves is too much for two owners.
Keep the app as the place to start work and run the shop's common actions.
Use the Google Workspace we already pay for to write, calculate, email and meet.
Keep engineering knowledge and SOPs in the Git vault, and transactions in a proven ERP.
Other open-source companies mix their own products with specialist tools, including Google.
Test ERPNext for simplification; do not rebuild everything while deciding.
Status: proposal
Recommendation: keep Home, shrink its scope. Adopt “every task starts in the app: find it, see status, act on common actions; deep work opens the specialist tool.” Keep the phone-first experience and Google sign-in. Stop treating an external editor as a design failure.
Numbered owner decisions — proposed, not yet adopted
- Adopted 2026-10-04 (owner: "i ok with changing the rule to the recommended"). Replace “every click stays inside” with the rule above. Routine receiving, building, testing and printing should still finish inside Home.
- Adopted 2026-10-04. Keep Google for human documents, Sheets, Gmail, Calendar and Meet; the vault for engineering/SOPs; InvenTree and Kanboard for their current records. Give every fact one authoritative home.
- Adopted 2026-10-04 with one addition (owner: "a workflow that generates a PDF into inventory or into our app… start in the app, then modify the document in Google Drive"): Publish PDF attaches a numbered PDF snapshot to the linked app/ERP record. Shrink Documents to an object-linked index, Open in Google and useful PDF previews. Defer automated document releases, Forms pipelines and calendar synchronization until a real job requires them.
- Stop expanding the personal-note editor, collaboration/history/sharing engine, generic chat features and any proposed accounting/email/calendar replacement. Preserve existing useful features and data; retirement is a separate implementation decision.
- Keep the ERPNext trial from the October 1 consolidation plan. Judge it by complete shop workflows and custom code removed, before choosing any migration.
- First measure one week of real use; then prioritize one recurring shop bottleneck and one Documents pilot. Approve further features from saved time, fewer errors and maintenance cost.
Evidence: what companies actually do
Research checked October 4, 2026. These are public self-reports, not audits of live tenants. Older evidence is dated below; “unknown” means not established by the sources reviewed, not “they do not use it.” This sample includes open-source vendors, remote companies and hardware businesses; they are not all fully open-source or small.
Each entry covers knowledge/documents, tasks, ERP/finance and Google where disclosed. Product capabilities do not prove a company uses them internally. Vendor customer stories have promotional bias.
GitLab — handbook plus several specialist systems. Durable knowledge lives in its Markdown handbook; Google Docs supports collaboration, with lasting guidance moved into the handbook. Work uses GitLab issues/MRs. Google Workspace is explicitly supported; NetSuite appears in financial access controls. Communication policy, Markdown guide, supported apps, July 2026, financial controls.
GitLab's June 2026 Glean guide describes search across handbook, issues, Google Workspace and other systems while keeping content in those systems. That supports a shared discovery layer; it does not justify buying Glean for two people. GitLab's own explanation.
PostHog — a particularly close match for the proposed split. Its handbook is in GitHub; issues/PRs hold product work. Google Docs holds internal meeting notes and updates. Onboarding lists Google Suite, including Docs/Sheets/Slides, plus Slack, Salesforce and Fondo for US accounting; a manufacturing ERP is unknown. Tool inventory, communication rules.
PostHog explicitly avoids drafting handbook-bound material in Google because that duplicates effort and leaves the handbook stale. Its own onboarding portal coexists with specialist services. This is evidence for bounded custom workflows, not a universal custom application. Same communication policy.
Sourcegraph — do not mistake its old Git handbook for today's complete stack. A currently indexed company page names Notion for handbook/team wikis; full-page retrieval failed. Historical Git handbook files specify Google Docs RFCs and GitHub/Asana work artifacts. Current task system, ERP and Workspace subscription are unverified; historical Google Docs use is explicit. Current handbook search result's source, historical RFC policy, historical workflow.
Oxide Computer — Git knowledge does not exclude Google. Engineering RFDs are AsciiDoc in GitHub, discussed through PRs and presented through a reading interface. Its September 2026 incident policy names Google shared Drive and Google Docs as the incident record, with Workspace administration. Company-wide tasks and ERP are unknown. RFD tooling, 2023, incident system of record, 2026.
37signals/Basecamp — the strongest centralization counterexample. It reports 98% of internal communication in Basecamp, including project work, policy updates and decisions. This is product dogfooding with a maintained commercial tool. Separate office-document, ERP and Google Workspace arrangements are unknown here; “98%” concerns communication, not all operations. Company communication guide.
Frappe — an integrated family of products, not literally one app. Its September 2023 account places accounting/orders/inventory in ERPNext, policies/SOPs in Frappe Wiki and discussion plus some tasks in Gameplan. It then used Google Drive and planned replacement by Frappe Drive; current Google use is unverified. First-party operations account, current Wiki dogfooding statement.
Odoo user: Assembly Technologies. Odoo's August 2025 case study describes a 75-person electronics manufacturer replacing Sage/Arena and adopting sales, purchasing, inventory, manufacturing, accounting, PLM and Documents. Its handbook, task authority and Google use are undisclosed. This demonstrates suite consolidation at a larger manufacturer, not a free two-person implementation. Customer case.
Nextcloud GmbH — documented dogfooding, with limits to what is public. Its 2022 account explicitly locates the company handbook in Collectives. That establishes an internal knowledge home; the reviewed evidence does not establish the complete document/task/ERP stack or whether it also uses Google Workspace. Do not infer “no Google” from its product positioning. Company handbook example.
Automattic — P2 knowledge plus other tools. P2s hold persistent discussion and institutional knowledge; Slack and Zoom/Meet serve other communication. Special Projects explicitly uses Google Docs/Sheets for some authoritative project documents. GitHub notifications also appear in staff expectations; a universal task system, ERP and exact Workspace deployment are unknown. Communication, Google document practice, work expectations.
Zulip — dogfood conversation; keep development work in GitHub. Its public development community uses Zulip, while GitHub issues/PRs and zulipbot organize contribution work. Developer knowledge lives in published project documentation. Internal business documents, ERP and Google Workspace use remain unknown. Community, documented GitHub workflow.
Mattermost — own chat alongside Google and business systems. Its handbook directs staff to Mattermost for internal communication and Google Docs/Sheets templates for documents. It has a published handbook; HR documentation mentions NetSuite financial reporting and shared Google Drive. A single company-wide task authority is not established here. Google tools are confirmed; exact Workspace edition is unknown. Staff FAQ, HR systems.
Hardware comparisons — useful, but not two-owner peers
SparkFun: its 2016 report describes the internally built Sparkle ERP and a software development team maintaining stock-location/cycle-count tools. Current ERP, internal knowledge/documents, tasks and Google Workspace are unknown. The report also describes performance problems following the change: custom control carries operating work. Sparkle report.
Adafruit: a 2019 company job post describes custom PLM, ERP and ecommerce, supported by four internal developers plus outside developers. Internal knowledge/documents, task systems, current ERP architecture and Google Workspace are unknown. This is historical evidence for specialized manufacturing software with dedicated staffing. Adafruit's job description.
Prusa: ABRA's historical customer account says Prusa outgrew Google spreadsheets and adopted ABRA Flexi accounting in 2016/17, integrating its shop and a separate warehouse system through APIs. Current knowledge/docs, tasks, full ERP and Workspace deployment are unknown. Google spreadsheet use is historical, not proof of today's subscription. Vendor account quoting Josef Průša, Czech.
Framework: its current company directory lists a NetSuite engineer. Inference: NetSuite is an internal competency, but the title does not establish which records it owns. Internal knowledge/documents, tasks, complete ERP deployment and Google Workspace use remain unverified. Company directory.
What these patterns mean for EmbeddedEra
| Strategy | Trade-off for this shop |
|---|---|
| Custom “everything inside” app | Consistent screens; we maintain every copied workflow. |
| Specialist tools + thin Home | Shared entry, search and common actions; occasional handoffs. Recommend now. |
| ERPNext/Odoo suite + small Home | Fewer transactional boundaries; migration and configuration work. Trial before adopting. |
Handbook-first is a writing discipline: record a decision in its durable home, then link it from conversations. It does not require every document to be Markdown or every screen to belong to one application. GitLab and PostHog provide direct examples above.
“Integrate, don't replicate” means reuse the engine's rules and authoritative records. A custom screen over an API still creates maintenance: permissions, validation, upstream changes, phone behavior and recovery. A thin hub can grow into a super-app unless its common actions remain deliberately limited.
Cost evidence: app-notes-editing-plan estimated 19–28 focused engineering days for the whole editor plan; notes-options-google-vs-own estimated another 10–16 days for missing core features after the live pilot. These are local estimates, not measured costs or additive totals. The app release log records repeated phone fixes and a test-isolation failure. This is already a maintained product. embeddedera-app
Google's SRE guidance explains why recurring operational work expands without active limits. It does not establish a two-person shop's maintenance percentage. Inference for us: AI can reduce drafting/coding time, but owners still carry integration, recovery and review; dependence on one person's understanding creates a bus factor of one even with many agents. Operational toil.
The app's benefits remain valuable: one familiar phone entry, common sign-in, fewer repeated searches, coordinated shop actions and visible agent work. Specialist apps can still require authorization or account handoff. Agents need stable APIs and permissions, not all tools rendered inside one browser page. Measure maintenance hours alongside time saved; $0 licence cost is not $0 operating cost.
What Home should own, and what it should link
| Home owns | Authoritative engine remains |
|---|---|
| Shop-specific receive/build/ship screens | ERP stock, parts, BOMs and transactions |
| Unit history, test/firmware links and labels | ERP serials; tester evidence; existing print services |
| Agent status, action receipts and approvals where required | Existing jobs, policies and domain permissions |
| One actionable inbox and notifications | Linked task, failure or approval record |
| Permission-aware search and object links | Original ERP, vault or Drive record |
| Phone navigation and common actions | Existing authentication and service APIs |
“Own” means own the experience and missing glue, not invent a second stock ledger or reproduce every ERP feature. Prefer native InvenTree/ERPNext functions and maintained plugins before custom shop logic. Reports remain views over their sources; reuse Metabase instead of building a BI product.
Link/index: Google documents and spreadsheets; Gmail, Calendar and Meet; accounting in the chosen maintained accounting system; deep document editing, complex reporting and uncommon administration in their specialist tools. Home may show relevant totals/status and launch a specific record. An invoice screen is not a general ledger.
Keep the vault: engineering decisions, SOP sources, project context and the ICM server map. Assigned task status stays in Kanboard until a proven replacement exists; notes link to it instead of creating another task authority. Index only records the user may see, label their source and freshness, and keep search caches disposable.
Etsy/POD: retain the existing shop platform, fulfilment integrations and approved-art records. Home should expose blocked listings, order exceptions and agent outcomes. Do not turn the electronics ERP into a second Etsy order editor or apply board-build workflows to outsourced POD.
Stop adding: a rich personal-note editor, collaborative merge/diff UI, generic file-sharing platform, full chat competitor, mail/calendar clients, spreadsheet engine, accounting engine and screen-for-screen copies of specialist tools. Keep working chat/notes features stable until an actual replacement and recoverable transition exist. No rewrite simply to undo sunk cost.
Google Documents: keep the direction, shrink the first delivery
The newer google-workspace-documents-plan already accepts opening Google for editing and records the owner's preference for Google for human writing. Keep that boundary; do not revive the older quick-capture/editor expansion. Business Starter is the supplied company baseline; this research did not inspect the tenant or billing.
Start with one Documents list: title, source, linked part/build/project/customer, and Open. Link existing files first; add PDF preview where it saves a real phone handoff. Create/share/rename/history can initially happen in Google. A document link grants no access; Google's permissions remain authoritative.
Use the existing company-owned Drive arrangement. Keep personal writing private until shared, and keep company originals under company control. Delay a new folder bureaucracy, multi-drive release register, form-response importer, calendar sync and agent editing identities until their specific need appears. No new subscription is recommended.
Defer the automated release workflow, not necessary production control. When an instruction actually governs a build, retain the approved PDF, approver/date/revision and build link through the existing controlled process. Never substitute a changing draft for required approved instructions. Automate releases only after one real workflow demonstrates the need; the full prior release design remains a later option.
For API previews, Google login alone does not authorize private Drive access; use narrow per-user OAuth and the maintained client. Opening the editor stays the default. ERPNext also documents OAuth for its Google integrations. Drive authorization, ERPNext Google settings.
ERPNext: a bounded trial, not an assumed migration
The September mrp-tool decision favored InvenTree; the later October 1 plex-server consolidation log explicitly proposes an ERPNext trial to decide the engine. The ICM plan also proposes proven engines behind Home. Its move map is a separate reorganization plan, not evidence that ERP migration succeeded.
Keep today's sources authoritative during evaluation. Trial one real assembly in isolation: supplier part → PO/receipt → BOM/build-to-stock → serial/test evidence → label → finished stock → Paul Shop consignment → shipment/invoice → return. Include revisions, existing IPNs, actual cost, approvals and both owners' phone use.
Compare native workflows before building custom screens. Pass only if the trial reduces total custom work while preserving inventory identities/history, tester/printing integrations and recovery. Include accounting fit with the company's accountant when that work is resumed. An API demonstration alone is insufficient; price implementation, upgrades and migration effort.
ERPNext documents Google Calendar/Contacts integrations and Drive integration, but the documented Drive feature uploads ERP backups. It is not a Google Docs editor or a complete document index. Google and the Git vault can remain even if ERPNext wins. Integration settings, Drive's actual scope.
Odoo is a credible suite alternative, but its own August 2026 comparison places advanced manufacturing/quality capabilities in Enterprise. Do not assume a customer's Odoo Online deployment is reproducible with Community alone; community add-ons require fit and maintenance checks. No paid switch is justified by this research. Edition comparison.
If ERPNext passes, migrate deliberately and retire overlapping adapters/tools after reconciliation. If it fails, keep InvenTree plus the smaller hub. Avoid maintaining two writable inventories or replacing Kanboard with a homemade database merely to reduce the menu count.
Rules of thumb for future requests
- Does a maintained tool already do this? Configure it, link it, then build only missing glue.
- Is the pain recurring and shop-specific? Name the real job and demonstrate the gap before coding.
- Where is the original? One authoritative record; everything else is a link, cache or view.
- Can opening the specialist tool solve it? Try that before recreating its interface.
- Can an agent use a supported API? Prefer that; an API key is convenient, but OAuth is valid when required.
- Who repairs and restores it if the main maintainer is away? Keep recovery instructions and a specialist-tool fallback.
- Does this save more owner time than it consumes in maintenance? Count errors and interruptions, not feature totals.
- Can we add it without another subscription? Use free/community or already-paid tools unless a demonstrated gap justifies more.
Next review: after a week, compare owner minutes for receiving, finding an instruction and clearing an agent exception; record failures and maintenance time too. Expand only the demonstrated bottleneck. A shared front door succeeds when work is easier, regardless of which editor opens.
Prepared by the primary Codex session; no helper models ran (capacity evidence missing; read-only helper confinement unavailable). Only this proposal was written. The explicit one-file scope excludes sync, commits, memory updates, code changes and implementation.
10-line summary
Keep Home as the company front door.
Replace “every click stays” with “every task starts.”
Build only the shop-specific workflow gaps.
Let the ERP own inventory and transactions.
Use already-paid Google for human office work.
Keep engineering knowledge and SOPs in Git.
Start Documents with links and useful previews.
Defer release automation until a real build needs it.
Trial ERPNext for simpler operations before migrating.
Judge success by owner time saved and fewer errors.