Both www.discuss.io and app.discuss.io are already fronted by Cloudflare today. The next move is making the layer below — video at scale, Genie Experience Agents, the Insights Agent, and the Voxco + Ascribe integration plane — run on the same edge.
A quick architectural read from the public internet — useful as the starting point, because Discuss is already a Cloudflare customer for both the marketing site and the application. The expansion question becomes the developer platform underneath, not the edge in front.
wp.wpenginepowered.com) · Cloudflare in front (server: cloudflare, cf-cache-status: HIT)ec2-52-6-12-229.compute-1.amazonaws.com) · Cloudflare in front (cf-team, cf-ray)ns-*.awsdns-*.org/com/net/co.uk) — same NS set for both apex and subsdefault-src https:, HSTS with includeSubdomains, connect-src * — opportunity for tighter policy at the edge without app changesThe picture is consistent with a fast-moving SaaS that picked Cloudflare for the front door and AWS for the application tier. The interesting frontier is the layer Discuss is actively building now: Genie Experience Agents, Discuss Now (AI-led interviews), the Insights Agent, and the Voxco + Ascribe combination. Every one of those has a direct mapping to a Cloudflare developer primitive.
The platform page at discuss.io/platform describes the lifecycle as Prepare → Ask → Analyze, with Genie Experience Agents working across all three. Here is the same lifecycle, with the Cloudflare primitive that satisfies each stage's runtime requirements.
Every other line in Discuss's value proposition gets simpler when video storage and delivery costs do. Today the workflow lands video into AWS, transcodes it there, and serves it back to analyze — paying egress on every step.
R2 has zero egress to the public internet. Move the qual repository to R2 once, and every Insights Agent retrieval, every highlight reel, every customer download stops accruing data-transfer charges.
Live virtual focus groups, self-paced video open-ends, and recorded IDIs share one delivery path. Stream handles encoding, DRM, signed URLs, and global delivery from the same anycast edge.
Embed every transcript on ingest. The Insights Agent answers "show me every time a Mondelez consumer mentioned chocolate-substitute friction" across the entire archive — not just the current study.
A single Discuss Now interview routes through STT, LLM reasoning, and TTS. The Insights Agent then runs another pass for theming, sentiment, and highlight extraction. That is four-to-six model calls per respondent — and Discuss serves 50,000 hours of interviews per year.
Cloudflare shipped dollar-denominated spend limits and identity-driven budgets in AI Gateway this month. For Discuss, the mapping is direct: every Genie Experience Agent, every Insights Agent run, every Discuss Now session can route through a single inspection plane that enforces per-study, per-customer, per-agent cost ceilings before any upstream model call leaves Cloudflare's network.
Combined with caching, fallback routing, prompt logging, PII redaction, and per-route rate limits, this gives Discuss a single audit log answering the regulator's question about which model saw which respondent transcript when. For 100+ markets of PII-laden consumer interviews, that audit surface is not a nice-to-have.
PepsiCo's three concurrent studies and Mondelez's brand-tracker should have separate, dollar-denominated budgets — not a shared API key whose end-of-month invoice nobody can explain.
AI Gateway attributes spend by model, provider, or any custom attribute Discuss sends — study ID, customer ID, Genie instance ID, moderator ID — in real time.
Multi-stage Genie agents in retry or recursion patterns are the canonical "code review bot burned through 50M tokens" failure mode from Cloudflare's own engineering.
A dollar-denominated cap at the gateway tier ends the loop before the bill arrives, and can fall back to a cheaper model instead of just refusing the request.
Consumer interview transcripts contain names, locations, brand preferences, household composition, and often health information. GDPR Article 6 + 9 and CCPA both apply.
AI Gateway DLP rules can strip PII and secrets before the request leaves Cloudflare for OpenAI, Anthropic, or Google — and log redaction events to R2 for the audit trail.
Discuss Now likely calls one model for STT, another for the conversational interview, a third for live theming, a fourth for highlight extraction. Today that is four invoices and four dashboards.
AI Gateway gives one unified log with per-provider breakdown, so finance can answer the per-study cost question without a custom dashboard build.
Discuss recruits respondents from over 100 markets. A French consumer interview must respect French data residency. A German respondent's transcript is GDPR Article 9 special-category data the moment they mention health. A California respondent triggers CCPA. The current single us-east-1 application tier cannot honor those constraints on a per-respondent basis.
Cloudflare Regional Services guarantees TLS termination and processing inside a customer-selected jurisdiction. EU respondents stay in EU regions; US respondents stay in US regions; APAC follows the same rule.
No application change. The same Cloudflare edge that already fronts app.discuss.io takes on the residency role.
Some Discuss customers — PepsiCo, Mondelez, Pernod Ricard — already have data-handling commitments to their own end-consumers and regulators. The current product stops at "Discuss has logs."
Logpush writes audit telemetry directly into a customer-owned R2 or S3 bucket. Chain of custody intact. Discuss becomes the platform that lets the customer own their own audit story.
No rip-and-replace. Discuss is already on Cloudflare; the work is turning on the developer-platform primitives below the edge. Every phase is reversible by toggling a route.
Route one Genie agent's calls through AI Gateway. Per-study spend caps, PII redaction, fallback routing. Benchmark cost-per-study against the current path. Hand finance a real per-study P&L.
Migrate one large-volume customer (Danone-scale) to R2 + Stream. Measure egress savings, transcoding throughput, retrieval latency from the Insights Agent. Open a benchmark conversation with engineering.
Embed one quarter of one customer's transcripts into Vectorize. Wire it into the Insights Agent's retrieval layer. Demo the cross-study question: "what have we ever heard from this segment about this category?"
Discuss was named one of only two Leaders in The Forrester Wave™: Experience Research Platforms, Q1 2026. The Voxco + Ascribe strategic combination expands the surface area meaningfully. Genie Experience Agents are now a publicly stated pillar of the platform. The infrastructure underneath needs to keep up with all three at once.
The fastest-moving AI-native companies — Anthropic, ElevenLabs, Perplexity — chose Cloudflare for the layer underneath their product, not just the edge in front of it. Discuss already made the first half of that decision. The second half is on the table now, with AI Gateway, R2, Stream, Vectorize, Workers AI, and Realtime all GA and battle-tested at the scale Discuss is approaching.