Floom product beats — L08–L12

Independent critique of the 5 product beats (the approved white-card animated versions) by Gemini 2.5 Pro (frames + voiceover + intent) and Codex (creative director, reasoning from descriptions). The white-card look is the approved base; every note is a way to make each beat stronger.

L08Establishing6/10
L09Spawn6/10
L10Schedules & triggers4/10
L11Secured & sandboxed5/10
L12Model-agnostic · open source6/10
Gemini average across the five beats: 5.4/10. Read: a good base, every beat has a clear path up.
L08
Establishing
“A platform for autonomous background agents that actually work.”
planned (establishing beat — no change planned)
Gemini 2.5 Pro · multimodal 6/10
What works
  • The core visual language is established and adheres to the style guide. The orb, cards, and color palette are on-brand and clean.
  • The central metaphor of an 'Orchestrator' with orbiting 'agents' is clear and easy to grasp from the visuals provided.
  • The 3D orb itself is a high-quality asset. The lighting, refraction, and soft glow make it a compelling focal point.
Improvements
  1. The composition is too static and centered. In the late frame, the entire element group feels like a flat diagram. Shift the orb and its cards left-of-center to create more dynamic negative space and a less predictable layout.
  2. The 'orbit' is unconvincing. The cards just fade-and-rise into place along a static path. To sell the idea of a dynamic system, the cards should physically travel along the orbital path into their final position, not just appear there.
  3. The iconography on the cards is inconsistent and confusing. The mix of dots and a checkmark in the late frame feels arbitrary. To reinforce the VO '...that actually work', have all cards animate to a 'completed' checkmark state simultaneously on the word 'work'.
Codex creative director
What works
  • The orb as a recurring orchestrator anchors the product world and gives the system a recognizable character.
  • The three job cards quickly communicate plurality: Floom is not one chatbot, it is a platform running multiple workers.
Improvements
  1. Rename `THE ORCHESTRATOR` to `FLOOM ORCHESTRATOR`. The current label is cinematic but slightly abstract; this beat needs product ownership.
  2. Make each orbiting card carry one tiny proof of autonomy: `scheduled`, `drafting`, `complete`. That turns “cards around orb” into “background agents doing work.”
  3. Keep the ellipse faint, but make the cards rise one-by-one before orbiting. The first 1.2s needs legibility before motion beauty.
L09
Spawn
“You can spawn agents with a single prompt.”
planned add agent/provider logos (Claude Code · Codex · Cursor · ChatGPT) by the prompt box
Gemini 2.5 Pro · multimodal 6/10
What works
  • The core aesthetic is intact. The clean cards, color palette, and type choices align with the approved direction. It feels premium, not generic.
  • The basic card animation is clean. The rise-in with motion blur and opacity, as seen in the mid-frame, reads as a deliberate, smooth transition rather than a rendering glitch.
  • Information hierarchy on the cards is logical. The prompt is clearly separate from the result, and the key metrics on the result card are prominent and easy to parse.
Improvements
  1. The 'spawn' is a lie. The VO says 'spawn an agent' but the visual shows a finished report card just sliding into frame. This is a massive missed opportunity. Instead of a simple rise, the prompt card needs to emit something—a particle, a thread of light—that visibly travels and *constructs* the new card. Show the magic of creation, don't just show the receipt.
  2. The UI is unfinished and lazy. Those gray placeholder lines on the right of the 'Signups Digest' card are unacceptable for a final render. It looks like a low-fidelity wireframe. Replace this immediately with a crisp, animated sparkline chart that actually visualizes the '+18%' data point. Don't ship mockups.
  3. The composition is static and disconnected. The prompt and result cards just sit near each other. Create a stronger visual link. When the 'Run' button is notionally clicked, have it pulse blue, then draw a subtle, glowing line that connects the prompt to the result card as it forms. This visually reinforces the 'single prompt' cause-and-effect narrative instead of relying on proximity alone.
Codex creative director
What works
  • Prompt in, agent out is exactly the right causal idea for this beat.
  • The result card has useful metrics, which makes the spawned agent feel productive instead of decorative.
Improvements
  1. The planned provider logos are right, but place them as a compact “spawn from” rail above or beside the prompt: `Claude Code`, `Codex`, `Cursor`, `ChatGPT`. Do not let them compete with the prompt/result causality.
  2. Change the result label from `spawned from your prompt` to `agent created`. Shorter, more premium, clearer.
  3. Time the beat as three clean hits: prompt appears on “single prompt,” Run pulses once, result card rises on “spawn agents.” Avoid extra micro-animations; this is a clarity beat.
L10
Schedules & triggers
“They run in the background, ask for approval, and wake on schedules or triggers.”
planned show scheduling as a real cron card
Gemini 2.5 Pro · multimodal 4/10
What works
  • The foundational aesthetic is correct. The color palette, typography, and soft shadow card style align with the 'cool, restrained' brief.
  • The staggered rise-in animation of the cards is fundamentally sound. The motion blur and opacity ramp seen in the mid-frame create a clean, sequential reveal.
  • The 3D glass orb remains a strong, consistent visual anchor, successfully grounding the composition on the left.
Improvements
  1. Fix the layout immediately. The final approval card collides with the founder PIP in the late frame. This is a critical compositional failure. Either raise the entire card stack or reduce the vertical margin between cards to ensure no overlap. Premium work does not have elements crashing into each other.
  2. The card UI is weak and lacks the specified 'Vercel-grade' detail. The 'Every weekday' card's blue dots are meaningless; replace them with a concrete UI for a schedule, like a mini-calendar view or a cron-like syntax. The `* trigger` tags are tacked on; integrate this information more elegantly. The approval card's orb icon is redundant next to the main orb; use explicit 'Approve' / 'Deny' buttons to make the 'approval' function clear.
  3. The timing is wrong. The 'approval' card animates in while the VO discusses 'schedules or triggers'. This breaks the narrative link. Re-time the animation: the approval card must land precisely on the 'ask for approval' VO line. The schedule and trigger cards should then follow on their respective VO cues. The current timing is confusing and sloppy.
Codex creative director
What works
  • This beat covers the strongest product promise: agents do not just run once, they live in the background.
  • Approval is present, which adds trust and keeps autonomy from feeling reckless.
Improvements
  1. The planned CRON card is correct. Use concrete copy: `CRON · 0 9 * * 1-5` with a human line below: `Every weekday at 09:00`. That satisfies builders and remains readable.
  2. Split the beat into three visual chapters: `Background`, `Triggers`, `Approval`. At 8.5s, the current version risks becoming a cluster of cards unless the staging has hierarchy.
  3. Make the approval card the final and largest card: `Approve 14 drafted replies?` with two actions, `Review` and `Approve`. The story lands better when control is the final note.
L11
Secured & sandboxed
“Every run is logged, secure, and sandboxed.”
planned show sandboxing as its own distinct card
Gemini 2.5 Pro · multimodal 5/10
What works
  • The visual metaphor of the orb inside the translucent ring with a shield icon is a clear and effective representation of 'secure and sandboxed'. It's the strongest part of the beat.
  • The overall composition is clean and balanced. It adheres to the established grid without feeling cluttered, maintaining the premium aesthetic.
  • The 'RUN LOG' card itself is legible and the basic information hierarchy is understandable, even if the execution is uninspired.
Improvements
  1. The visuals are disconnected from the VO's cadence. Instead of the ring and shield appearing simultaneously as seen in the mid-frame, they should animate in sequentially to match the words 'secure' and 'sandboxed' respectively. This creates a tighter, more satisfying sync.
  2. The 'RUN LOG' card is static. This is motion design, not a slide deck. Animate the log lines typing out with a cursor, and have the checkmarks at the end of each line animate in with a distinct 'tick' motion. Show a process, don't just display the result.
  3. The card's visual details are weak and miss the mark for a premium feel. The grey checkmarks in the late frame are nearly invisible; make them the brand's accent blue (#3E6FE0) and slightly larger. The 'secured' pill just sits there; it should animate in with a pulse or glow after the final log line completes to provide a definitive 'complete and secure' moment.
Codex creative director
What works
  • The run log is strong proof. It makes “logged” feel real instead of a compliance buzzword.
  • The containment ring around the orb gives sandboxing a visual metaphor that fits the approved look.
Improvements
  1. The planned separate sandbox card is necessary. Use two cards: `RUN LOG` for auditability, `SANDBOX` for isolation. Copy: `Ephemeral workspace`, `Scoped tools`, `No ambient access`.
  2. Remove or reduce the shield under the orb if the sandbox card enters. Shield plus ring plus secured pill can become security icon soup.
  3. Animate the log as a precise terminal-like fill: one line per VO word group. `run started`, `sandbox provisioned`, `tool approved`, `run complete`. Keep it calm and exact.
L12
Model-agnostic · open source
“Zero lock-in. Fully model-agnostic and open source.”
planned providers + GitHub — make the open-source / model-agnostic proof land harder
Gemini 2.5 Pro · multimodal 6/10
What works
  • The visual storytelling is direct and effective. The provider logos clearly communicate 'model-agnostic' and the GitHub card shows 'open-source', perfectly matching the voiceover.
  • The '0 lock-in' card design is strong. Using the large '0' as a graphic element is a good hook and makes the benefit immediately scannable.
  • The staggered animation of the logos followed by the main cards is consistent with the film's established motion language and creates a clean, readable build.
Improvements
  1. The provider logos in the late frame look randomly placed and cheapen the composition. Animate them into a structured grid or have them briefly orbit the blue orb before settling. This will make them feel like an integrated system, not just floating clipart.
  2. The GitHub card lacks credibility and looks like a template. The '.. star' in the late frame is an immediate red flag. Replace it with a plausible high number (e.g., '1.2k') or remove the count and just show the 'Star' button. The sub-text is also illegible; replace it with clear, iconic tags like 'MIT License'.
  3. The two main cards feel disconnected. The VO says 'and', but the visuals are just two separate columns. After they land, add a subtle connecting animation—a pulse from the orb that highlights both, or a thin line that draws from one to the other—to visually link 'no lock-in' with its 'open source' proof.
Codex creative director
What works
  • Providers plus GitHub is the right proof stack for “zero lock-in.”
  • The GitHub card gives the open-source claim a concrete destination instead of leaving it as copy.
Improvements
  1. Make the provider row unmistakably model providers: group only `OpenAI`, `Anthropic`, `Google Gemini`, `Meta Llama`, plus maybe `local`. Gmail reads as an integration, not a model provider, so move it out or remove it here.
  2. Replace `0 lock-in · switch models any time` with `Zero lock-in` as the headline and `Switch models, hosts, and clients any time` as the support line. It broadens the claim without getting wordy.
  3. Make the GitHub proof land harder with repository UI language: `floomhq/floom`, `MIT`, `Self-host`, `Deploy anywhere`. The star button is fine, but the license and self-host tags carry more credibility.