Inside the AI-Augmented Workflow at My Studio (the Boring Useful Version)

The previous post on AI tools that actually moved the needle covered which tools earned a spot. This one covers how they're plugged together. Less hype, more "here's the boring stack, here's where it lives in the workflow, here's what's still hands-on."

The shape of the stack matters because the breathless AI-replaces-everything pitch and the dismissive AI-is-just-hype pitch both miss the actual story: AI is most useful when it's invisible inside an otherwise normal production workflow. Not the centerpiece. Just another tool on the truck.

The agent layer (OpenClaw)

The studio runs an agent layer called OpenClaw on top of the core production stack. In practical terms, this means: Slack, Discord, email, calendar, and file-system events all flow through an agent that can answer, route, draft, summarize, schedule, and act on them with explicit approval. It's the connective tissue between "human knows the work" and "AI handles the parts that don't need human judgment."

Where this shows up in client work:

• Inbox triage. The agent reads the day's email, surfaces what actually needs an answer, drafts replies in the studio voice, and waits for a one-word approval before sending. Time saved per week: easily 4–6 hours.
• Calendar coordination. Shoot scheduling, edit deadlines, client meetings — all routed through the agent, which knows the studio's actual calendar shape and won't propose a conflict.
• Content drafts. Blog posts (this one was drafted with agent help), case study skeletons, social caption variations, email outreach templates. Drafts go in, human edits go on, ship.
• Cross-channel routing. Client message in Slack. Operational task in Discord. Calendar event blocking the studio. The agent reconciles them without the human having to manually check four surfaces.

The non-obvious part: the agent is more useful for what it prevents than for what it produces. Missed messages, scheduling overlaps, things falling through cracks — gone. The cost of context-switching between channels has dropped to near zero.

The visual gen pipeline (Krea)

All image and video generation routes through Krea — a unified API that brokers across the major model providers (Kling, Hailuo, Seedance, gpt-image, nano-banana, ideogram, Flux Kontext, and dozens more) without the studio needing to maintain accounts or quotas with each upstream provider individually.

Why route through one platform instead of going direct:

• Single billing surface. One credit pool, one history, one audit trail. Don't have to reconcile five provider invoices to see what a project actually cost.
• Model abstraction. When a new model releases, it's a one-line change to use it. The workflow doesn't care whether Kling 2.5 or Hailuo 2.3 generated a clip — both come back as a URL.
• Three-direction discipline. Every visual asset for a client gets generated in three different model lanes (Cinematic, Action Hero, Graphic Poster) so the client sees a real spread, not a single take to argue about. This is way easier when the platform supports all three directions natively.

What the studio uses Krea for in practice:

• Concept-stage moodboards and visual directions for brand campaigns
• Subtle motion on static brand assets (the homepage avatar is a Kling 2.5 image-to-video output)
• Social-content motion graphics that don't need a full After Effects timeline
• Pitch decks where placeholder visuals would otherwise eat days of a designer's calendar
• Quick-turn editorial content for time-sensitive campaigns

What the studio still does the hard way: the actual hero deliverable. Final commercials, final case-study video, final brand spots — those are shot, edited, and graded by humans on real cameras. Krea accelerates the path to those deliverables; it doesn't replace them.

Voice and transcription

The voiceover work on owned channels is still mine. AI voice cloning got tested for the brand voice work and got dropped — see the previous post for why.

Where AI voice does earn its keep: transcription. Whisper-class models running locally (or in the cloud through the same agent layer) handle interview transcription, edit-pass subtitle generation, and rough metadata for archive search. The voice generation lane is closed for client deliverables. The voice recognition lane is wide open and saves hours per project.

Edit-room AI (selective)

The edit room uses AI selectively — auto-cut to a music track for a v0 assembly, auto-pull selects from interview footage, auto-arrange B-roll to a narrative arc. None of these are final-cut quality. All of them save real hours getting to the first watchable rough cut, which is when the actual creative editorial work begins.

The pattern is consistent: AI for the v0, human for the v1+. The first watchable assembly happens 50–70% faster. The actual taste, pacing, and story decisions are still done by an editor who knows what the brand actually sounds like.

What the integrated stack looks like in practice

Take a typical mid-week shoot day for a multi-property brand client:

1. Morning: Agent has the calendar pre-blocked for the shoot. Shot list and brand brief pulled into the day's working file. Crew confirmed via overnight Slack triage.
2. On-site: Capture happens on actual cameras with actual humans. Pocket 3 for run-and-gun, R5C for hero shots. No AI on the camera side.
3. Edit-in-trailer: Footage uploads, transcription runs in the background, v0 assembly drafts in editing software while the human starts the real edit pass. AI is preparing the room while the editor does the work.
4. Motion / graphics: Lower-thirds and brand title cards from a Krea-generated palette feed the After Effects pipeline. Final motion is rendered in AE; the visual ideation came from Krea.
5. Voiceover: Recorded human voice. (No AI clone for brand voice work.)
6. Delivery: Files render to spec for each surface (broadcast, social vertical, web embed). Agent handles file naming, client-portal upload, and the "delivered" Slack message.

The human is doing the hard work. The agent is removing every form of friction that isn't the hard work. That's the entire pitch.

Why I think this is where the market lands

Studios that try to AI-replace the actual creative work are going to ship forgettable content and lose to studios that don't. Studios that refuse to integrate AI into pre-production and post-production are going to lose to studios that ship in half the time.

The middle path — AI as connective tissue, AI as v0-accelerator, AI as ideation surface, human as the last mile — is going to look like the obvious move in three years. Right now it's still a competitive advantage because most working studios haven't built the stack.

If you're a peer producer reading this, the question I'd start with isn't "what AI tool should I try?" It's "what part of my workflow eats my calendar that has nothing to do with the actual creative work?" That's where the agent layer earns its keep. The visual-gen and editorial-AI tools come second.

If you're a brand or operator reading this, the question is simpler: does the studio I'm working with show up faster, more consistent, and more responsive than the last one? The stack is one explanation for why the answer is yes here. The other explanation is the human keeps showing up.

Talk about what this looks like for your project →

← Back to all posts · See case studies