Team roles for an AI creative reverse-engineering pipeline

TL;DR: Four core roles run the workflow regardless of team size: strategist (picks references and directs audience-fit), deconstructor (does the 10-layer visual analysis), caster (handles template casting + image prompts + copy), QA (validates output before media handoff). A 1-person team has one person wearing all four hats; a 10-person team has dedicated specialists per role plus coordination. This cluster defines each role concretely, maps role combinations to team sizes, and explains the handoff protocol between them.

The four core roles

Every AI creative reverse-engineering pipeline runs through these four functions, whether they’re four separate people, one person doing all of it, or an agent orchestrated by a human.

Role 1 — Strategist

What they do:

• Pick which competitors to monitor
• Pick which references to reverse-engineer from the current week’s candidates
• Define the audience-fit parameters (what audience are we targeting; does the reference match)
• Approve the reverse-engineering go/no-go per reference
• Set the testing hypothesis for the variation set
• Make escalation calls on legally or brand-sensitive decisions

Skills: performance marketing judgment, audience understanding, category knowledge, business context awareness.

Failure mode when absent: the team reverse-engineers every available reference regardless of fit. Output volume is high but performance is flat because many ads weren’t strategically chosen.

Role 2 — Deconstructor

What they do:

• Execute the 10-layer deconstruction on the reference
• Abstract the structural formula from competitor-specific details
• Produce the template specification with enough precision to be executable
• Flag structural choices that won’t transfer cleanly to the user’s product

Skills: visual analysis, art direction literacy, eye for lighting / composition / framing archetypes.

Failure mode when absent: the team casts from fuzzy deconstructions that miss structural details. Output lands as surface mimicry rather than structural mimicry.

Role 3 — Caster (production)

What they do:

• Execute template casting — swap skin, preserve structure
• Generate image prompts in the target model’s native prompt style
• Write native-language copy per variation
• Produce the 5-variation set with hypotheses
• Handle the AI-tool interaction (Midjourney, Flux, DALL-E, Nano Banana, Ideogram, etc.)

Skills: prompt engineering, copywriting, native-language competence, knowledge of image-model behaviors.

Failure mode when absent: generic AI-slop output. Creative that technically matches the template but feels like every other AI ad.

Role 4 — QA

What they do:

• Validate output against the quality-check list
• Verify trademark-safety and regulatory compliance
• Check character limits on copy across all languages
• Confirm the 5-variation set has meaningfully distinct hypotheses (not variance for variance’s sake)
• Flag output that needs human review before media handoff

Skills: attention to detail, policy knowledge, quality judgment, willingness to kill bad creative.

Failure mode when absent: bad creative ships. Character-limit overruns, translation errors, trademark risk, uncanny-valley AI images all make it to paid media.

Team sizes and role combinations

1-person team: the polymath operator

One person wears all four hats, sequentially per ad. Realistic output: 10–15 reverse-engineered ads per month.

Pattern: Monday — strategy day (picks refs, sets hypotheses for the week). Tuesday–Thursday — deconstruction + casting per ad. Friday — QA and media handoff.

Strengths: no handoff overhead, full context per ad. Weaknesses: limited by one person’s breadth and velocity; bad week means zero output.

When this scales: Primores-sized consulting engagements, small DTC founder-operators, growing creative teams pre-hire.

2-person team: strategist + production lead

The strategist does roles 1 + 4 (strategy + QA). A production lead does roles 2 + 3 (deconstruction + casting).

Output: 20–30 ads per month. The strategist sets direction Monday; the production lead executes through the week; weekly Friday review for QA and handoff.

Strengths: natural separation between “what to do” and “how to execute” — reduces scope creep, sharpens focus. Weaknesses: still fragile — one person out for a week cuts output in half.

3–5-person team: specialized production + part-time strategy

Strategist role can be part-time (the head of marketing or growth lead spends 20% of their time on this). Dedicated deconstructor, dedicated caster (prompt + image), dedicated copywriter (or caster + copywriter combined). QA is either a senior team member or split across the team in peer review.

Output: 40–60 ads per month.

Strengths: depth per role, parallelization possible. Weaknesses: coordination overhead; requires explicit handoff protocol (see below).

6–10-person team: full creative ops function

Strategy becomes a dedicated role (a creative ops lead or senior strategist). Deconstruction may be split across product specialists (one person knows fashion, another knows beverage, another knows supplements). Casters specialize by platform (Meta vs TikTok vs Pinterest). Copywriters specialize by market/language. Dedicated QA person.

Output: 80+ ads per month across multiple brands / campaigns / markets.

Strengths: can run multiple parallel workflows, maintain quality at high volume, handle complexity (multi-market localization, regulated categories). Weaknesses: significant coordination cost; needs clear processes to avoid becoming an agency-style creative department.

Handoff protocol between roles

In any team larger than 1 person, explicit handoffs prevent context loss. The minimum handoff artifact per step:

• Strategist → Deconstructor: reference ad URL or screenshot, audience-fit notes, go/no-go, hypothesis direction.
• Deconstructor → Caster: the template specification (10 layers), any structural elements flagged as “won’t transfer cleanly,” the abstracted formula.
• Caster → QA: the full 5-variation set (image prompts, copy, hypotheses), any open questions about trademark / compliance / native-language confidence.
• QA → Media: approved creative with variant-level hypotheses, any flagged risks, character-limit validation per language.

Primores’ internal workflow codifies these as structured JSON artifacts that pass between roles (or between phases of the ad-alchemy skill when one person is orchestrating). Structured handoff is what makes the workflow skill-executable — without it, every handoff becomes a verbal conversation.

Where AI fits per role

AI tools don’t replace any of these roles — but they make each role 3–5x more productive:

• Strategist: AI helps monitor competitors at scale (monitoring workflow), surface patterns across many ads, draft audience-fit analyses. Doesn’t replace strategic judgment.
• Deconstructor: multimodal models execute the 10-layer analysis in 10 minutes that would take a human 45–60 minutes. Doesn’t replace the human’s final calibration of what transfers.
• Caster: prompt generation + copy drafting at scale. Language models can write native-level copy in 30+ languages when given the template. Image models execute the prompts. Human still selects which output to use.
• QA: LLMs can run the first-pass check against the quality-check list; humans do final sign-off especially for trademark / brand / cultural sensitivity.

The right mental model: AI is the force multiplier inside each role. Team-sizing advice above assumes AI is in the loop; without AI tools, the roles still exist but velocity drops 3–5x.

Hiring signals per role

For teams building out:

• Strategist: performance marketing background (3+ years), comfort with category research, willingness to say no to references that don’t fit. Marketing manager or senior growth marketer profile.
• Deconstructor: art direction background or strong visual analysis skills. Often a former designer or photographer. Can also be a creative strategist if they’re calibrated on visual craft.
• Caster: hands-on AI tool experience + copywriting background. Rare combo — often hired as “creative technologist” or “prompt engineer + writer.” Native-language competence in target markets.
• QA: senior team member with strong attention to detail and willingness to kill bad creative. Often the head of creative ops or senior editor. This role gets more important as team size grows.

Key takeaways

• Four roles: strategist, deconstructor, caster, QA. Constant across team sizes.
• 1-person team wears all four hats sequentially. 10–15 ads/month.
• 2-person team = strategist + production lead. 20–30 ads/month.
• 3–5-person team = specialized production with part-time strategy. 40–60 ads/month.
• 6–10-person team = full creative-ops function. 80+ ads/month.
• Explicit handoff protocol is what makes larger teams work at scale.
• AI is the force multiplier inside each role, not a replacement for any role.

Related

• seo/ai-creative-reverse-engineering-complete-methodology — the pillar
• ai-template-casting-workflow — the caster’s workflow
• competitor-ad-monitoring-workflow — the strategist’s monitoring layer
• glossary/visual-deconstruction — the deconstructor’s framework
• scaling-ad-creative-volume — how team size maps to output capacity
• glossary/ai-creative-reverse-engineering — canonical definition

Sources

• Primores internal creative-ops observations across small-team (2-person) to mid-sized (5-person) client engagements, 2025–2026.
• Creative-ops structures documented in performance-marketing industry practice (Motion, MNTN, Ample).