Planned blog posts
One teaser per topic from the 100-post map. No outlines here — just the queue.
C1 — Category: why “agents” is the wrong noun (demand: scarce OS shelf)
- 1Why “AI agents” became wallpaper (and what replaces the noun)published
- 2The operating-system claim: rare, useful, easy to fakepublished
- 3Six lenses of a company OS (memory, execution, workforce, cockpit, infra, vertical)planned
- 4Company-agnostic vs vertical OS — same engine, different shapeplanned
- 5Boards are read-models, not the runtimepublished
- 6Observe → truth → propose → gate → humans decide (the loop as product spine)published
- 7Control plane vs chat plane: where the product actually livesplanned
- 8Category succession stories (CRM → ERP → next) without buying the metaphorplanned
- 9Anti-pilot language that doesn’t lieplanned
- 10What “corporate agentic OS” must refuse to beplanned
C2 — Observe: nothing typed twice (demand: channels + docs; Stage-1 C: T03)
- 11Company observation as a first-class product jobplanned
- 12Email as the real system of record (and why scanners die)planned
- 13Multi-channel ingress without a second inboxplanned
- 14Capture portals that don’t become another boardplanned
- 15Document intake is observation, not “DMS AI”draft
- 16From PDF to event: parsing that must survive policyplanned
- 17Dedup, UID traps, and why “synced” ≠ “true”planned
- 18Observations vs opinions: what is allowed into the ledgerplanned
- 19Invisible demand: counting requests nobody loggeddraft
- 20The observation anti-pattern: type moreplanned
C3 — Truth: case, memory, graph (demand: brain/ontology 42; Stage-1 C: T04,T44,T38)
- 21Shared operational truth vs chat contextpublished
- 22The Case object: durable situation, not a threadpublished
- 23Ledger first: why projections come lastplanned
- 24Company memory that isn’t a vector dumpplanned
- 25Ontology / living graph without “company brain” mushplanned
- 26OCED-style case events: the minimum event shapeplanned
- 27Decision lineage: who approved what, on which factsplanned
- 28Approved baseline ≠ current timeline (the only honest slip)published
- 29Citations that bind a claim to a source documentplanned
- 30When the graph is marketing chromeplanned
C4 — Understand & predict (demand: anti-pilot 36; criteria 4–5)
- 31Causal understanding vs “what is late”planned
- 32Future understanding: what breaks nextplanned
- 33Earliest controllable source of largest avoidable consequenceplanned
- 34Lead-time exposure before site delaydraft
- 35Intervention pre-assessment: should we even act?planned
- 36Process mining for companies that don’t process-mineplanned
- 37Agent behavior mining / span mapping as honesty about workplanned
- 38Trace maturity: when a span is evidenceplanned
- 39Collective intelligence without a second meetingplanned
- 40Prediction that is falsifiable (or it isn’t prediction)planned
C5 — Gate & authority (demand: HITL 27 + govern 35; Stage-1 C: T13,T16,T34)
- 41Autonomy ↔ human authority: name both sidespublished
- 42HITL as a runtime primitive, not a FAQpublished
- 43Propose → gate → execute as UI statesplanned
- 44Policy gates: agents execute, policies decideplanned
- 45Approval calibration: not every approval deserves a humanplanned
- 46Stalled approval ≠ slow approvalpublished
- 47Exception-only surfaces: only the exception appearsplanned
- 48Emergency control and kill switches that are realplanned
- 49Access vs authority (RBAC is not a gate)planned
- 50Audit trails that name prompt, tools, output, approverplanned
C6 — Execute & orchestration (demand: orchestration 39; C02/C05)
- 51Closed execution loop vs recommendation dashboardpublished
- 52Durable gated execution: survive the restartplanned
- 53Workflow orchestration that isn’t Monday with an LLMplanned
- 54Tool calling without turning the company into an API zooplanned
- 55Sandboxes: what agents may touchplanned
- 56Skill registry: capabilities as declared inventoryplanned
- 57Self-coding agents: power with a maturity gateplanned
- 58Assistant ↔ expert delegation (humans stay in judgment)planned
- 59Digital workers vs operating layer — two GTMs, one homepage riskplanned
- 60No order without lead time and schedule consequenceplanned
C7 — Verify, learn, sleep (demand: eval stacks; Stage-1 enablers)
- 61Verify outcome ≠ HTTP 200draft
- 62Eval harnesses for operational agentsplanned
- 63Critic / evaluator in the loopplanned
- 64Observability that explains *why*, not only latencyplanned
- 65Self-monitoring without another dashboard farmplanned
- 66Self-learning loops that don’t silently rewrite policyplanned
- 67Sleep / offline improvement as a first-class jobplanned
- 68Importance consolidation: what to keep, what to forgetplanned
- 69Guardrails that bind action, not just textplanned
- 70Earned autonomy: levels you can measureplanned
C8 — Interfaces people actually see (demand: market defaults to chat; our gap)
- 71Chat is the default product — and why that’s a trapdraft
- 72Four planes: employee, builder, runtime, governanceplanned
- 73Surface taxonomy: what must exist before “AI UI”planned
- 74Front page as projection of cases, exposures, waiting decisionsdraft
- 75Builder studios vs declared company shapeplanned
- 76Governance chrome that appears in the mock, not the footerplanned
- 77Wireframes for the minimum OS (what to draw first)planned
- 78Criteria click-paths: if you can’t click it, you don’t have itplanned
- 79Construction-vertical chrome vs Case+Gate (steal carefully)planned
- 80Shown interfaces vs claimed interfaces (C0–C3 honesty)planned
C9 — Documents as the observe substrate (36-vendor shelf)
- 81AI document management: six market categories, one OS jobplanned
- 82IDP that feeds a ledger, not another folder treeplanned
- 83Contracts as obligations with dates and consequencesplanned
- 84Clause search for AEC without becoming a law-firm DMSplanned
- 85Metadata-first filing vs chat-over-filesplanned
- 86Agentic document workflows: approve before ERP writeplanned
- 87Multimodal parse infra for agents (not a second product)planned
- 88Knowledge graphs from documents without losing the sourceplanned
- 89Vaults agents can access under policy (MCP-native memory)planned
- 90Document anti-list: what not to promise on a DMS landingplanned
C10 — Market language & GTM (demand: how to speak without mush)
- 91Five verb bets: Govern, Run, Work, Build, Knowdraft
- 92Hero formulas that still work (and the ones that don’t)planned
- 93Guide of attention: belief order after the H1planned
- 94Category-creation moves that survive a viewport checkplanned
- 95CTA pairs: demoplanned
- 96Anti-patterns library: platform-for-agents and friendsplanned
- 97Peer shelf for an agentic OS (who is actually adjacent)planned
- 98Measurement language over feature gridsdraft
- 99Self-serve agent marketplace vs declared company-shape OSplanned
- 100The publishing spine: how these 100 posts reinforce one thesisdraft