Product Feature Research 2026-04-28 · v0.1 · 5 competitors · 126 pages
Product

Feature matrix, UX prescriptions, & the IA decision.

The 50-row matrix maps where each competitor stands. The CreateOS playbook strip below points at the prescriptive moves; the UX section tells you which 7 patterns to adopt and which 5 levers to differentiate on.

Playbook · Feature matrix · UX optimization

The CreateOS playbook — four objectives, four top moves

Each card distills CreateOS's highest-leverage move per objective, with the anti-patterns competitors fell into. Full playbook pages linked below each card.
Acquisition
Ship the no-signup demo at a dedicated short domain — "try it in 60 seconds, claim it later."
Do — CreateOS-specific
  • Build a public-facing demo where a user types → live app in 60s → shareable URL, no signup. Claim-this-project via email-only.
  • Acquire a short, verbal domain for the keystone experience (create.new tier). $10-100K, outsized return.
  • Target OpenAI Agents SDK + Anthropic integrations + LangChain. One SDK inclusion = a year of content marketing.
  • Weekly founder-led public metrics post on X/LinkedIn. Tuesday cadence. Start Week 2.
Avoid
  • Competing head-on with Lovable on prompt-to-app. They have 18-month head start.
  • Marketing all 14 frameworks equally. Pick one demo stack; let the rest be a footnote.
  • "Create account" as the only CTA above the fold. It pre-filters out browsers.
Read the Acquisition playbook→
Retention
Ship per-agent spend caps + persistent usage widget before MPP mainnet scale.
Do — CreateOS-specific
  • Persistent usage widget in sidebar — projects used, agent credits, MPP transactions, creator earnings.
  • Per-agent spend caps + budget isolation + kill-switch before any mainnet launch. Replit's July-2025 DB deletion is your cautionary tale.
  • Visible agent reasoning (named stages OR tool-call logs, pick one). In-flight stop button.
  • Version history with auto conventional-commit labels + visual diff on scrub.
Avoid
  • Letting autonomous agents run with unbounded spend. Incident blast radius grows with MPP volume.
  • Hidden credit consumption — surprise bills churn users 3-5× faster.
  • Ephemeral-feeling projects. Credits contracting MoM means projects aren't being revisited.
Read the Retention playbook→
Revenue
Reposition MPP as x402-aligned. Productize the agent-specific pricing tier.
Do — CreateOS-specific
  • Publish MPP ↔ x402 alignment within 30 days — facilitator, extension, or compatible. Pick one, commit publicly. Register with x402 Foundation.
  • Launch the agent-economy pricing tier: configurable per-agent caps + budget isolation + MPP dashboard. The feature gap that unlocks enterprise.
  • "Made with CreateOS" watermark + credit bribe on free-tier deployments. Turn every earning agent into distribution.
  • Creator revenue share visible + marketable — public leaderboard of top marketplace creators with earnings.
Avoid
  • Positioning as rival protocol to x402. Coinbase + CF + Google + Visa is too much gravity to fight.
  • Enterprise pull toward "help us own compute." Wrong conversation. Route to NodeOps / partners.
  • Heavy reliance on flat subscription for heavy agent users — margin compression kills you (see Lovable: $400M ARR, not yet profitable).
Read the Revenue playbook→
Strategy
Agent Economy Week in Month 2-3, not Month 10. Window measured in months, not quarters.
Do — CreateOS-specific
  • Your primitive: MPP + router.nodeops + marketplace. Assembled, live, differentiated. Present as the combination, not a feature list.
  • Your narrowing: "Deploy an earning agent" keystone. Use primitives only you have. Stop fighting Lovable on their axis.
  • Your distribution event: Cloudflare-Agents-Week-style 5-day blitz. Keynote + cases + ecosystem + product + deep-dive. Book it.
  • BD Tier 1 (30 days): Karan Sharma (OpenAI Agents SDK), Harrison Chase (LangChain), Sri Viswanath (Sycamore), x402 Foundation.
Avoid
  • Feature parity with Replit / Lovable / Bolt. You have assets they don't; fight on those.
  • Deferring integrations to Phase 3 (the original playbook error). Accelerate to 30 days.
  • Hiring a growth team before PLG is working. Amjad publicly said this backfired at Replit.
Read the Strategic Bets playbook→
The hardest truth: CreateOS has accumulated more primitives than most competitors had at their breakout. But the category is consolidating in real time — and without a narrow positioning + a launch event in the next 90 days, Cloudflare absorbs the agent-economy narrative.
CreateOS brief 30-60-90 plan Full playbook

Feature matrix

Side-by-side. Cells link to the underlying wiki pages.
Feature Replit Anything.com Fly.io Hercules Multica Canonical
AI & Agent capabilitiesCore generation + autonomy patterns
AI app generation agent
✓Agent 4 — versioned, self-verifies via screenshot
✓Conversational Builder — GPT-5, 40+ integrations
◐Phoenix.new — Elixir/Phoenix only, narrow community beachhead
✓Chat → app → deploy. Non-tech-founder focus.
—Not a builder
 Fly scoped to one ecosystem.
Parallel agent execution (Task Board)
✓Task Board kanban (Core for new tasks)
—Not productized
—Not a builder
?Not captured
◐Multi-agent assignment via board
 Replit-only.
Autonomous build / test loop
✓Autonomous Build Mode — Core-gated
✓Max — paid add-on, contextual ("Use MAX here")
—Not a builder
?Not captured
◐Agents auto-execute assigned tasks
 Both builders. Packaging differs.
Plan-before-execute mode
✓Plan Mode — toggle on prompt (Core)
◐Reasoning blocks inlined, no explicit review gate
—Not a builder
?Not captured
—Not a builder
Background / async task execution
✓Background Tasks — Core
?Max behavior when tab closed unknown
✓Machines designed for long-running / always-on workloads
?Not captured
✓Full lifecycle (enqueue → claim → start → complete/fail)
 Fly wins on infra semantics.
Cost/quality mode dial
✓Economy pill on chat bar
✓Auto pill on prompt (persistent)
—No LLM UI
?Not captured
—User picks the agent CLI
 Builder-only pattern.
Custom instructions (persistent)
?Not captured
✓Per-project (AI Settings); team-shared library referenced
—No LLM UI
?Not captured
◐Per-agent instructions when creating an agent
Named design-system presets
—None captured
✓"High-Fidelity SaaS", "Ghost Structure" — agent snaps outputs
—Not a builder
?Not captured
—Not a builder
 Hidden moat for Anything.
Agent runtime primitivesAdjacent infra: where the agent actually executes
Hardware-isolated sandbox (agent runtime)
—Own stack, not exposed
—Own stack, not exposed
✓Sprites — "Run AI-generated code in complete safety", dedicated CPU/memory/network per instance
—Hosted apps, no sandbox surface
◐Local daemon + cloud runtimes; not VM-isolated
 Fly's headline AI play.
GPU compute
—Not offered
—Not offered
✓Fly GPUs — per-second billing, top-level docs
—Not offered
—User brings own compute
 Infra-only capability.
microVM / per-second compute billing
—Flat bundle pricing
—Credit metered, not compute metered
✓Fly Machines — hardware-virt'd, per-second, 40% reservation discount
—Tier-bundled hosting
—User pays own LLM + compute
 Different pricing paradigm.
Output artifactsWhat the agent can ship
Web apps
✓Full web app output
✓Sites + apps (Next.js)
✓Any framework via fly launch
✓SaaS, eCommerce, Internal Tools, Landing
—Not a builder
Mobile apps (iOS + Android)
✓Native + Core Mobile Simulators
✓Via Expo Go preview client
—No mobile-build tooling
✓App Store / Google Play deploy claimed
—N/A
 Builders only.
Landing pages
✓Distinct artifact on template carousel
—Not surfaced
—Not a surface
✓Distinct artifact + example chip
—N/A
Video / Slides / Animation / 3D / Data viz
✓All five advertised as templates
—Out of scope
—Out of scope
—Out of scope
—Out of scope
 Replit breadth claim.
Visual design canvas
✓Infinite Canvas — spatial whiteboard, peer to preview
—Code editor only
—Not a builder
?Not captured
—Issue-tracker UI, not canvas
 Replit-only.
Figma import (on prompt)
✓First-party docs page, named time
✓Figma icon on prompt toolbar — not marketed
—Not a builder
?Not captured
—Not a builder
Full code editor (in-browser)
✓First-class, always visible
✓Next.js file tree + JSX editor — not marketed
◐Only in Phoenix.new — not in Fly console
?Not captured
—Not a builder; assigns to coding-agent CLIs
Version history
✓Task-level versions, "Main version" radio
✓Auto conventional-commit labels (feat: …)
◐User's Git, not a product surface
◐Built-in pillar; depth unknown
◐User's git via the agent's coding-tool
Platform servicesBackend primitives bundled with the product
Authentication (platform-level)
✓Built-in, zero setup
?Not visible in captures
✓Fly SSO / OIDC / tokens
✓Built-in pillar
✓Workspace auth + tokens
 Fly = platform login only.
Auth providers (for built app's users)
◐Via integrations
✓Auth Providers — dedicated project settings
—User's responsibility
✓'Users & Auth' pillar
—N/A
 Anything separates the two concerns.
Database
✓Built-in, hidden in bundle
✓Top-level dashboard nav
✓Managed Postgres + Tigris (S3) + Upstash (Redis)
✓'Database' pillar
✓Postgres 17 + pgvector
 IA rank differs.
Hosting + custom domains
✓Encrypted + custom domains
✓"Domains" in dashboard nav; custom domain Pro-gated
✓Core product — Machines + anycast + TLS
✓'Domains' + hosting pillar
—Self-host or Multica Cloud — not app hosting
 Fly sells this directly.
Monitoring
✓Built-in service
?Not mentioned
✓Metrics + logging + error tracking built-in
✓'Analytics' pillar
✓Per-runtime usage + activity heatmap
Secrets / env vars
✓Legacy first-class feature
✓Project Settings → API Settings
✓flyctl secrets — CLI-native
?Likely yes
✓API keys held on user's machine
Native payment acceptance
◐Stripe integration (1 of 100+)
✓"Accept payments" top-level + Monetization settings
—Not a Fly concern
✓'Payments' pillar (built-in commerce)
—N/A
 Sharpest builder-side differentiator.
App Store marketing / ASO
—Not attempted
✓Dedicated settings page (Mobile App)
—Out of scope
◐App Store / Play deploy; ASO not detailed
—N/A
 Maker-audience moat.
Monetization config (subscriptions/IAP)
—Out of product scope
✓Mobile App Settings → Monetization
—Out of scope
◐Implied via Payments pillar
—N/A
Social share / OG cards
?Not surfaced
✓Share Settings → Social Share
—Out of scope
?Not captured
—N/A
Integration marketplace
✓100+; Connectors is the Core-tier surface
✓"40+ integrations" claimed on landing (GPT-5 named)
—Different category — no marketplace
✓1000+ APIs claimed; not catalogued
◐10+ coding-agent CLIs as 'providers'
Templates / remix
✓Community-sourced (remix) + carousel on home
✓Templates (likely curated)
◐Framework quickstarts (Phoenix/Rails/…) — not a remix gallery
✓'From the Community' categorized gallery
—Not a builder gallery
Developer surfaceHow much of a "dev tool" this is
Command palette (⌘K)
✓Account / team / org / trash / clear
—Not captured
?Not captured — console UX not in raw/
?Not captured
?Not captured
 Dev-audience signal.
Mobile simulators in-browser
✓Core feature — iOS + Android
—Expo Go on physical device only
—Out of scope
?Not captured
—Not a builder
 Infra-heavy moat.
Mobile app as build surface
✓Native iOS + Android app
—Web-only build
—Out of scope
—Web-only build surface
—No mobile build
Persistent usage / credit widget
✓Always visible in sidebar (3 meters)
—"Get free credits" CTA only
?Dashboard UX not captured
?Not captured
✓Per-runtime token spend (Input/Output/Cache R/W)
 Transparency-by-default.
Invite / collaboration
✓Project + workspace + account seats
✓Project-level invite
✓Organizations + members (flyctl)
?Not captured
✓Multi-workspace, member roles
 Replit richest hierarchy.
Trust & complianceSecurity posture + enterprise-readiness
Security scanning (CVE)
✓Core-gated; scan history, CVE severities
—No security posture captured
—User's responsibility
?Not captured
—User's responsibility
SBOM (Software Bill of Materials)
✓Downloadable reports (Core)
—No equivalent
—No equivalent
—Not advertised
—OSS — auditable, no SBOM artifact
 Replit-only B2B moat.
SOC 2 / GDPR / SSO/SAML
✓Type 2 audited, EU-ready, Enterprise SSO
—None signalled
◐SSO + OIDC + HIPAA add-on ($99/mo); SOC 2 not captured
?Not captured
◐Self-host = compliance answer
Customer logos on homepage
✓Google, Microsoft, Stripe, PayPal, SoFi, Zillow, Plaid, Gusto
—None
—None captured
◐'Trusted by 100k+ users' (count, no logos)
—None
 Replit's enterprise signal.
Workspace / Account / User settings hierarchy
✓3-level nesting; account seats
—Single-workspace IA
◐Org + member, no project-level
?Not captured
✓Multi-workspace + member roles
Publish + growthGet the app live + get more users
Publish flow scope
✓Per-artifact (pick one) — rich config
✓Project-wide with Web/Mobile split
✓fly deploy — whole app, rolling
✓'One-click publishing' claimed
—Not a publisher
Auto-deploy on push
✓Toggle in publish panel
?Not surfaced
✓GitHub Actions integration documented
?Not captured
◐Issue assignment is the trigger, not git-push
"Made with X" watermark / referral loop
✓Made with Replit — +50 credits per click; removable via upgrade
—No equivalent
—No equivalent
?Not captured
—Not advertised
 Replit-only viral loop.
In-app feedback widget
✓Core-gated
—Not offered
—Not offered
?Not captured
—Not advertised
Competitive migration imports (named)
✓GitHub, Figma, Bolt, Lovable — in docs
—Figma only (silent); no Bolt/Lovable
—No "import from Heroku / Render / Railway" play
—None advertised
—Not a builder
 Replit's aggressive capture play.
Creator education content
?Not captured
✓"Anything University" on YouTube
✓Phoenix Files, Laravel Bytes, Ruby Dispatch, Django Beats, JS Journal
?Skills nav item — possibly
◐OSS contrib model, no formal University
 Fly runs 5 per-stack pubs.
PricingTier structure + monetization
Free tier
✓Starter plan — 10 apps, credit-metered
✓Free credits
?Referenced but scope unclear
✓'Generous free hosting'
✓All features, self-host
Pricing model shape
✓Starter → Core (bundle) → Enterprise
✓Free → Pro → Max (add-on)
◐Pure metered — no tiers, calculator-only, 40% reservation discount
◐Tiered (specifics not captured)
◐OSS free; Cloud waitlist
 Fly = cloud pricing, not builder pricing.
Enterprise motion
✓"Contact sales" in nav; Security Scanning + SBOM in Core
—No enterprise motion
◐HIPAA $99/mo + support tiers + [email protected]; no logo wall
—None captured
◐Self-host as compliance answer
✓Present ◐Partial / via integration —Absent ?Unknown — need more sources

ICP coverage

Three buyer segments on the horizontal axis. Coverage shown as filled arcs — fuller arc = more direct targeting.
Non-coders Indie makers & solopreneurs Framework devs AI / agent builders Enterprise ring 1 · 70% ring 2 · 100% ring 3 · 80% primary · 100% ring 1 · 100% ring 2 · 70% ring 3 · 30% Replit Anything.com Fly.io
Replit three concentric rings

Broad, multi-audience bet — with real enterprise proof

Replit plays all three ICP rings simultaneously. Code editor + command palette + Library panel hold ring 2 (indie devs). Security Scanning, SBOM reports, 3-level settings hierarchy, and 8 enterprise logos on the homepage (Google, Microsoft, Stripe, PayPal…) win ring 3. Agent 4 + template carousel pull ring 1 (non-coders). The 2026-04-23 ingest sharpened this picture — Replit's ring 3 moat is procurement-grade, not marketing-grade.

Anything.com one focused ring

Sharp, single-audience bet — with a secret code editor

Every marketing decision serves indie makers / solopreneurs. Accept Payments is top-level; App Store Marketing, Monetization, and Named Design Systems are all maker-centric. But the screenshots revealed a full code editor, Figma import, and conventional-commit version history hiding underneath — features that contradict the "just describe what you want" positioning. Either strategic under-sell or relic of a pivot. Either way, the product is more capable than it admits.

Contested zone indie makers

The fight is still here, but converging

The UX shells look nearly identical now (left rail + three-pane + viewport toggles). Both have autonomous modes, both have mode dials, both have code editors. The competitive edge has shifted up the stack: Anything wins on monetization stack (payments + ASO + monetization config); Replit wins on power/platform bundle (Core unlocks Plan + Autonomous + Simulators + Background + Security in one step). Ring 1 users will pick based on how they feel about the maker features vs. the dev features.

Fly.io framework devs + AI/agent runtime

A different ICP axis entirely — dev-trust, not maker-economics

Fly owns framework-shipping developers (Phoenix, Rails, Django, Laravel, JS, Rust, Python, Go) via per-stack publications and CLI-first onboarding — that's ring 1. Sprites is the new ring 2: AI/agent builders who need isolated compute for generated code. Enterprise is a light ring 3 — a $99/mo HIPAA add-on and support plans, but no customer-logo wall. No indie-maker play at all. For CreateOS, the overlap with Replit/Anything is minimal; the overlap with Fly on ring 2 is direct and material. See Hardware-Isolated Sandboxes.

UI / UX optimization — what to adopt, what to differentiate, what to avoid

Prescriptions derived from 17 captured UX flows + Fly.io's CLI/SDK-first DX. Maps directly to CreateOS design decisions.

Replit and Anything.com have converged on the same builder shell — so copying it is free. The real UX leverage is one level deeper: inside specific components, and in a second discipline entirely — the agent-runtime / SDK-first DX that Fly.io (and the agent-framework listings) judge you on.

Discipline 1 · Builder shell (web)

Where users set up agents

The CreateOS dashboard / console. Three-pane workspace, prompt input, preview panel, settings hierarchy. Replit + Anything converged here — adopt the pattern, differentiate inside the components.

Lessons from: Replit (Agent Chat, Canvas, Task Board) · Anything.com (Conversational Builder, Project Settings, Max)
Discipline 2 · Agent runtime (API / CLI / SDK)

Where agents actually execute

The Sprites / E2B / Modal lane. No UI — just DX. Quickstart that fits on one screen, typed SDKs, framework examples (LangChain, OpenAI Agents SDK), per-stack quickstarts. Judged on "how fast can an agent make its first call."

Lessons from: Fly.io (fly launch, per-stack docs, per-ecosystem publications)

Adopt the converged shell — 7 patterns you don't need to reinvent

from playbook/04
Three-pane workspace
Left rail · main authoring · right preview. Collapsible.
→ pattern
Collapsible icon sidebar
~56px collapsed / ~260px expanded. Icons + labels.
Both competitors
Persistent top header
Viewport toggles + project name + Invite + Publish — consistent everywhere.
Both competitors
Prompt with mode pills
Attach · Figma · mode selector · submit. Single row.
→ pattern
Viewport toggle bar
Preview / code / 💻 / 📱 / ⟳ — icon bar in header.
→ pattern
Publish drawer
Right-side drawer OR dedicated tab — don't modal it.
→ pattern
Agent reasoning display
Named stages (makers) or tool-call logs (devs). Pick one, execute.
→ pattern
Don't fight the shell.
Both competitors converged — users expect these. Innovating on them is tax; innovating inside them is leverage.

Differentiate inside the shell — 5 levers to execute exceptionally well

components, not chrome
01
Named design-system presets
Anything's hidden moat ("High-Fidelity SaaS", "Ghost Structure"). Reduces taste burden + visual consistency. Make yours user-extendable → marketplace.
02
Cost-quality mode dial
Replit's Economy beats Anything's Auto — names the dimension. Show estimated cost per invocation at current setting.
03
Task board for agent orchestration
Replit's Drafts → Active → Ready → Done kanban. If you support parallel agents, operationalize the UI — don't leave it as plumbing.
04
Persistent usage widget
Replit's 3-meter left-sidebar widget. Always visible. Transparent-by-default. CreateOS adds MPP earnings as a 4th meter — competitor-unimitable.
05
Command palette (⌘K)
Low cost, outsized audience signal. Dev-adjacent users expect it; everyone benefits when they discover it. Replit ships it; Anything doesn't.
Fly.io lesson · The agent-runtime UX discipline

DX is the UX when there's no UI — and CreateOS is on this side of the line too

  • CLI-first onboarding. Fly's canonical first step is fly launch. CreateOS needs the equivalent for agents: npx createos init → sandbox + router + MPP wired in one command.
  • Per-framework quickstarts. Fly ships them per stack (Phoenix, Rails, Django, Laravel, JS, Rust, Python, Go). CreateOS needs them per agent framework: OpenAI Agents SDK, LangChain, LangGraph, Vercel AI SDK, raw TS / Python.
  • Per-ecosystem content publications. Fly runs 5 branded blogs (Phoenix Files, Laravel Bytes…). CreateOS analog: one ecosystem pub per major agent framework — Agents Dispatch, LangChain Field Notes — plus monthly earning-agent case studies.
  • Sandbox as API, not console. Fly's pitch is "agents run better on Sprites" — an SDK pitch, not a dashboard pitch. CreateOS must nail the SDK call shape before the console UI; the console is a convenience layer on top.
  • Docs index = product taxonomy. Fly's docs are organized by primitive (Machines / Sprites / GPUs / Postgres), not by use case. For infra-literate buyers this is clearer. CreateOS docs should mirror: Sandbox / Router / MPP / Marketplace — one top-level section each.
  • Calculator-first pricing fails for makers. Fly's metered pricing + calculator works for infra buyers but fails when the user can't estimate "what will one agent cost me." If CreateOS adds a maker-facing SKU on top of metered agent credits, publish the per-1000-actions rate, not just a calculator.
CreateOS-only UX unlocks · 5 moves no competitor can copy without entering your business

Where the research says you have unrepeatable surface area

  1. MPP earnings meter next to credits meter. Replit shows credits used; you show credits used and $ earned. First time a user sees a positive delta, retention spikes. No competitor has the payment rail to show this.
  2. Per-agent spend caps + budget bar. Surface a live budget bar per agent in the sidebar; when the cap is hit, the agent stops, not the account. This is the agent-era equivalent of Replit's SBOM — an enterprise procurement artifact no one else ships.
  3. Router telemetry surface. "Claude Haiku (0.3¢) · OpenAI GPT-5 (1.2¢) · local router decision" — showing per-call which model was chosen and why. Transparency moat against bundled builders whose router is a black box.
  4. "Made with CreateOS" on-chain attribution. Every MPP transaction can optionally include a micro-signature. Replit's watermark gives +50 credits; yours can give +% rev share. Turns every earning agent into distribution.
  5. Named agent presets. Anything's analog for agents: "research agent", "trading agent", "support agent" — each pre-wired to MPP + router + specific integrations. Lowers the zero-to-first-call friction the same way Anything's design presets lower the zero-to-first-frame friction.
  6. Earning-agent public feed. A public (opt-in) feed of anonymized agents currently earning — "Agent #4F2 earned $12.40 in the last hour." Social proof no one else can generate. Closest analog: GitHub's public commit feed. Competitors would need payment rails + agent primitives to replicate.

Anti-patterns — observed in the captured UX, do the opposite

7 traps pulled directly from the screenshots
  • Hiding features from marketing. Anything has a full code editor + Figma import — neither marketed. If you ship it, put it on the landing. Anything.com
  • Ambiguous mode labels. "Auto >" tells nothing; "Economy >" names the trade. Always name the dimension. Anything.com
  • Submit that doesn't flip to stop. A running agent needs a visible stop on the same pixel real-estate. Both do this; many don't. Both
  • Calculator-only pricing for builder-category buyers. Makers can't estimate "what will one app cost" from a calculator; publish rate cards. Fly.io
  • Hiding what the agent ran. Reasoning blocks OR tool calls — pick one. "Magical" without receipts = low trust. Both builders
  • Modal hell. Signup → onboarding → plan picker → first project → paywall. Progressive disclosure, not serial modals. Common
  • No in-product version diff. Users won't restore if they can't see the delta — version cards should re-render preview on select. Both builders
  • Flat settings page, 30 items. Anything's 6 sections (AI / General / API / User Auth / Mobile App / Share) are clean — emulate the discipline. Common

The one IA decision — your settings hierarchy is your audience statement

pick one — don't split the difference
IA shape Audience signal Who does this When CreateOS should pick it
Per-project sections
AI · Assets · Secrets · Auth · Mobile · Share		 Maker / solopreneur Anything.com If CreateOS adds a builder-facing product surface. Not the headline.
Workspace → Account → User
3-level nesting, enterprise multi-tenant		 Enterprise procurement Replit When you chase enterprise agent deployments (post Series A).
Org → Members → Tokens
flat, infra-flavor		 Developer / infra operator Fly.io The near-term default — matches agent-runtime ICP.
Agent → Task → MPP entry
new hierarchy — ships as CreateOS signature		 Agent economy CreateOS This is the bet. Invent the IA; make "agent" a top-level object with budget, MPP earnings, router telemetry, and audit log as peer tabs.
Full UX prescriptions + opinionated defaults (mode labels, shell, mode dial, version history format) live in the playbook.
→ playbook/04 — UX recommendations