analysis-areas.md
c8 — Analysis Areas (Draft for Review)
- 1. Source types (`SourceType`)
- 2. Use-case registry seed (`data/use-cases.json`)
- 3. Observation stream (`observations[]`)
- 4. Pain-point categories ⚑
- 5. Objection categories ⚑
- 6. Value-signal categories ⚑
- 7. Competitive-mention classes
- 8. Positioning-angle categories (which framings land)
- 9. Supporting vocabularies
- 10. Unchanged from phase 1
- 11. Open for review
Workstream B workshop document. Proposes the phase-2 vocabularies, the use-case registry seed, and the observation-stream design for Mark / Tyler / Cam review before any code changes. Sources:
inbox/qcontrol-as-signal-updated-text.md,briefs/endpoint-vs-network-objections.md,briefs/cam-dashboard-directives.md. Everything below is a draft — categories marked ⚑ are the ones I'm least sure about.
1. Source types (SourceType)
Who the feedback comes from. Tagged on companies (companies.json) with per-person overrides (known-entities.json).
| id | meaning |
|---|---|
build-partner | Active build-partner engagement (e.g. Cisco/Cam) |
prospect | Sales-motion conversation, no build engagement yet |
ciso-advisory | CISO advisory board / security advisors (incl. AMZ-style advisors external to prospects) |
investor | VC / investor conversations — now imported and tagged, not filtered out |
analyst | Industry analysts, press, category researchers |
industry-peer | Fellow founders/practitioners giving informal input |
internal | Qpoint team (AMZ's guidance stays internal-tagged, never prospect signal) |
2. Use-case registry seed (data/use-cases.json)
The named hypotheses extraction tests evidence against. Status: candidate → validating → validated / parked. First seed, drawn from Cam's directives + the positioning doc — needs Tyler/Cam review:
| id | hypothesis | initial status |
|---|---|---|
mcp-governance | Security teams need approval workflows over MCP servers/tools: configured-vs-connected visibility, tool manifests, redaction of unapproved tools before the agent sees them | validating (Cam: 6 SVP demos, real internal demand) |
inline-enforcement | Blocking unauthorized tool calls inline ("not authorized, don't invoke") with governed exceptions (expiry + audit trail) is the capability nothing else can replicate | validating (Cam's "agents at risk" panel reaction) |
sensitive-file-events | File-access visibility (credential reads, secrets, PII paths) surfaced as SIEM-routable security events — especially read→prompt→LLM-call connected sequences | validating (Cam: "that's what I wanna do") |
agent-inventory | Fleet-wide inventory of agents, sessions, MCP servers, and config-vs-runtime drift is the baseline security teams lack | candidate |
gateway-bypass-detection | Detecting agents bypassing the authorized AI gateway (direct-to-provider) and flagging them as at-risk | candidate |
heartbeat-signal | Endpoint heartbeat/attestation consumed by network enforcement — absence-as-signal, untrusted-by-default posture | candidate (core positioning, unproven with buyers) |
finops-context-waste | Quantifying idle MCP servers / context waste as inference cost — the FinOps side-door | candidate (Cam flagged as secondary) |
3. Observation stream (observations[])
One array per call; every item: type, category (per-type vocab below), text, verbatim, speaker, confidence, plus severity and productArea where relevant. Types declare workflow routing in vocabularies.ts.
Market-signal types (workflow: analyze)
| type | replaces | notes |
|---|---|---|
pain-point | painPoints[] | new category vocab §4 |
objection | objections[] | new category vocab §5 |
value-signal | valueSignals[] | new category vocab §6 |
competitive-mention | competitiveMentions[] | structured competitor class §7 |
budget-signal | urgency.budgetSignals[] | keep 5 phase-1 categories (positioning-neutral) |
risk | risks[] | keep phase-1 categories, review examples |
Product-feedback types (workflow: triage, rolled up in data/feedback.json)
| type | meaning |
|---|---|
bug | Something broken in the product they used |
feature-request | They need it (blocking or strongly wanted) |
feature-idea | Unprompted suggestion, brainstorm-grade |
concern | Worry about approach/direction (the advisory staple — e.g. "endpoint software is corruptible") |
usability-friction | Works but fights the user (install experience, dashboard confusion) |
deployment-blocker | Can't roll it out (MDM, OS coverage, VM/Docker gaps) |
praise | Specific positive on something real ("the file-read path is unique") |
Dual-workflow
use-case-evidence (routes to both): useCaseId, direction (validates / challenges / expands / new-variant), evidenceKind (usage-based / reaction-based / speculative).
4. Pain-point categories ⚑
agent-visibility-gap (can't see what agents do) · shadow-ai-usage (unsanctioned tools, personal devices) · block-vs-allow-binary (current options are all-or-nothing) · mcp-sprawl (unreviewed servers/tools loading into context) · agent-audit-compliance (can't prove agent behavior to auditors/execs) · prompt-injection-exposure (untrusted content reaching agents) · credential-data-access (agents touching secrets/sensitive files) · edr-network-blindspot (existing stack can't see local/agentic activity) · gateway-bypass (agents going direct around sanctioned paths) · agent-cost-drift (unmanaged inference cost, context waste, model drift) · incident-response (agent-related incident handling) · tool-fragmentation (point solutions not composing)
5. Objection categories ⚑
edr-overlap ("doesn't CrowdStrike/EDR already see this?") · gateway-overlap ("our LLM gateway covers it") · network-sufficiency ("we'll catch it at the perimeter") · endpoint-corruptibility ("endpoint software can be disabled/injected") · agent-fatigue ("not another agent on the endpoint") · coverage-gaps (VM/Docker/OS gaps, "multiply by 100 agents") · platform-vendor-wait ("Microsoft/CrowdStrike/Cisco will ship this") · performance-impact · positioning-clarity (can't place us in their mental map) · adoption-readiness (agentic usage too immature to secure yet) · security-trust (trusting a seed-stage vendor at this layer)
6. Value-signal categories ⚑
unique-visibility (sees what nothing else can — file reads, stdio MCP, local models) · inline-control (blocking before invocation) · signal-to-stack (feeds SIEM/network/gateway they already own) · exec-demand (leadership pull — "showed it to 6 SVPs") · immediate-use-case · deployment-appeal (endpoint model fits their environment) · pricing-positive · engagement-signals (unprompted follow-ups, internal championing)
7. Competitive-mention classes
edr (CrowdStrike, Defender, SentinelOne) · llm-gateway (portkey/litellm-class, internal gateways) · network-sase (perimeter, SASE, DNS-layer — incl. Cisco internal: Duo, Umbrella) · agentic-security (ASPM/ADR/AI-runtime-firewall startups) · platform-native (what Anthropic/OpenAI/Microsoft ship themselves) · diy (scripts, homegrown proxies)
8. Positioning-angle categories (which framings land)
endpoint-line-of-sight (sees intent+action in the same frame) · pre-network-visibility (before the packet exists) · signal-vs-enforcement (defense-in-depth handoff, ISE/Duo analogy) · heartbeat-attestation (absence-as-signal) · corrupted-agent-resilience (doesn't need agent cooperation) · continuity-not-restriction (enable AI use, don't block it) · gateway-complement (covers what the gateway can't see) · security-events-not-metrics (SIEM-routable events framing) · finops-angle (cost/waste side-door)
9. Supporting vocabularies
- Urgency drivers (revise):
agent-incident-triggered·board-ai-mandate·audit-compliance·internal-prototype-momentum(a Cam-style champion building demand) ·competitive-pressure·rollout-deadline - Product areas (feedback tagging):
agent-observation·mcp-instrumentation·policy-enforcement·heartbeat-attestation·inventory-discovery·dashboard·install-deploy·integrations-siem·docs-onboarding - Call types (additions):
working-session·build-feedback·deployment-debug·roadmap-planning·advisor-input(keep discovery/demo/technical-deep-dive/follow-up) - Personas (revise): split
security→ciso-security-leader/security-engineer; keep engineering-leader, devops-platform, developer, product, executive - New company-context fields (segmentation axes):
agentPosture(which agentic tools in use / sanctioned / banned),securityMaturity(SOC? SIEM? EDR vendor?),deploymentContext(managed endpoints, MDM, OS mix)
10. Unchanged from phase 1
Interest level 1–5 · funnel stages (prospect calls only) + engagement stage on company records · urgency block (budget/timeline/commitment) · decision-making block · value echoes (verbatim+speaker required) · key quotes · company size/stage/industry · themes (free tags).
11. Open for review
- Use-case list (§2): right hypotheses? right granularity? (Is
agent-inventoryits own use case or table stakes under all of them?) - SourceType (§1): does
ciso-advisoryvsindustry-peermatch how the advisory board actually works? - Objections (§5): match what Tyler/Cam hear live? Anything missing from recent calls?
finops-context-waste: worth tracking as a use case or fold into a value-signal category?- Prompt framing line to use everywhere: "Qpoint, a seed-stage company building endpoint security for AI agents — visibility, governance, and inline control over agentic tools on developer endpoints." (No "QControl" anywhere.)