Skip to main content

Buduję MVP-y i Claude Code automations dla solopreneurów. Od pomysłu do produkcji w 2-4 tygodnie.

Zarezerwuj 30-min audit (free) Zobacz co zbudowałem

Bartosz Gaca — MVPs, MCP Servers, Claude Code

Buduję MVP-y, MCP servery i Claude Code automations dla solopreneurów i tech-founderów. Od pomysłu do produkcji w 2-4 tygodnie. MVP Sprint 15-30K PLN, Builder Retainer 5-10K PLN/mies, Automation Pack 3-8K PLN/mies.

Usługi — od pomysłu do produkcji w 2–4 tygodnie

  • MVP Sprint — Twój produkt w produkcji w 2-4 tygodnie
  • Builder Retainer — Twój developer na abonament po MVP
  • Automation Pack — Claude Code + MCP + custom agenty
Zarezerwuj bezpłatny 30-min audit

Najczęstsze pytania

Ile trwa zbudowanie MVP?

Od pomysłu do produkcji w 2–4 tygodnie w modelu MVP Sprint — one-man software factory z Claude Code i custom agentami.

Ile kosztuje współpraca?

MVP Sprint 15 000–30 000 PLN jednorazowo, Builder Retainer 5 000–10 000 PLN/mies, Automation Pack 3 000–8 000 PLN/mies.

Co to jest MCP server i po co mi?

MCP server to most między modelem AI a Twoimi systemami (baza, API, narzędzia) — dzięki niemu asystent realnie wykonuje zadania w firmie, nie tylko odpowiada na pytania.

Dla kogo jest ta oferta?

Dla solopreneurów i tech-founderów oraz firm, które chcą automatyzacji i integracji AI bez budowania własnego zespołu deweloperskiego.

Jak zacząć współpracę?

Zarezerwuj bezpłatny 30-minutowy audyt na /audit — omawiamy zakres i najszybszą ścieżkę do produkcji.

I Built a 21-Agent Marketing Team in Claude Code — Content Without AI-Smell

After building a 19-agent dev team for software development, I built the same thing for marketing. 21 agents, 5 layers, 9 phases. Covers LinkedIn, Facebook, Instagram, TikTok, email, CRO and cold outreach. Never publishes without my approval — drafts only. This is an honest account after 2 months of use.

The problem with using one LLM for marketing

I asked Claude for a LinkedIn post. I got 200 words starting with "Excited to share that..." with emoji every two sentences. I asked for email copywriting. I got "Hi [First Name], I hope this email finds you well." I asked for a content strategy. I got a calendar with no connection to what I actually run.

The problem was not the model — it was the lack of context and specialization. Marketing covers dozens of sub-disciplines: brand storytelling is different from cold outreach, which is different from CRO, which is different from community management. One agent cannot handle all of it well simultaneously.

The fix was Mar Team v1.1: 21 agents, 5 layers, 9-phase pipeline, stored in ~/.claude/skills/team-mar/SKILL.md.

AI Marketing Team — 3 governance roles, 21 agents
Brand-first: Director / Architect / Cringe Checker + 4 execution categories.

Architecture: 5 layers, 21 agents

Layer 0 — Brand & Image (4 agents)

This is the foundation, initialized monthly, and it blocks anything off-brand from reaching production:

  • Brand Image Director — brand archetype, positioning, storytelling. Owns brand-identity.json
  • Brand Growth Architect — quarterly roadmap, growth channels, partnerships
  • Consistency Auditor — cross-platform brand consistency. Has veto power on off-brand content
  • Cringe Checker — detects AI-smell: "excited to share", "today's world", emoji every line. Also has veto power

Layer 1 — Strategy & Planning (2 agents)

  • Media Strategist (orchestrator) — content calendar, KPIs, coordination
  • Trend Scout — trending topics, viral content analysis, competitor watch

Layer 2 — Production (5 agents)

  • Copywriter — hooks, CTAs, A/B copy variants, tone per platform
  • Visual Designer — graphics, banners, carousels, infographics
  • Video Producer — reels, shorts, TikToks (ffmpeg + popcorn tools)
  • Content Repurposer — 1 post → 4 platforms (format, aspect ratio, hashtags)
  • SEO/Discovery Specialist — hashtags, keywords, alt text, schema markup

Layer 3 — Publishing & Analytics (5 agents)

LinkedIn Publisher, Facebook Publisher, Instagram Publisher, TikTok Publisher (placeholder — no TikTok API), Performance Analyst (cross-platform analytics, engagement rates, ROI).

Layer 4 — Growth & Retention (4 agents)

  • Email Marketer — lifecycle sequences (onboarding/activation/upgrade/win-back/churn)
  • CRO Specialist — funnel audits, A/B test hypotheses, paywall optimization
  • Community Manager — FB groups, Reddit, WhatsApp — contribution-first approach
  • Cold Outreach Specialist — ICP definition, prospect lists, 3-touch B2B sequences

Layer 5 — Bridge (1 agent)

Product-Marketing Bridge — dev↔mar sync: team-dev releases → content briefs; marketing asset requests → team-dev tasks.

9-phase pipeline

PhaseAgentsGoal
BBrand Image Director + Growth ArchitectMonthly brand brief
0Media StrategistStrategy — brand-aligned content calendar
1Trend Scout + CopywriterIdeation — hooks, angles, ideas
2Copywriter + Visual Designer + Video ProducerParallel production
3Content Repurposer + SEO Specialist1 content → 4 platforms
4Consistency Auditor → Cringe CheckerBrand review (both have veto)
5Publisher per platformDRAFTS ONLY — never auto-publish
6Performance AnalystEngagement analysis + recommendations
7Product-Marketing BridgeDev↔mar sync

Risk-based shortcuts:

  • QUICK POST (text only): 1 → 2 → 4 → 5
  • STANDARD (post + graphic): 0 → 1 → 2 → 3 → 4 → 5
  • CAMPAIGN (post series): all 9 phases
  • VIRAL RESPONSE: 1 → 2 → 4 → 5 → 6
  • FEATURE LAUNCH (post-deploy): 7 → 0 → 1 → 2 → 3 → 4 → 5 → 6

Cringe Checker — the AI-smell detector

This is the most interesting agent in the system. Cringe Checker holds a list of patterns (cringe-patterns.json) that mark content as AI-generated. Examples:

  • "Excited to share that..." — the classic LinkedIn AI hook
  • "today's fast-paced world..." — AI fluffy intro
  • "Game-changing", "cutting-edge", "revolutionary" — marketing BS
  • Emoji at the start of every bullet or every other sentence
  • "I hope this email finds you well" — cold email starters
  • Three or more adjectives before a single noun

When Cringe Checker flags content, it goes back to the Copywriter with specific reasoning. Not "this sounds like AI" — but "line 3: 'today's competitive landscape' is a typical AI-filler, replace with a concrete observation from your client's specific industry."

In practice: after the first iteration Cringe Checker catches 2-4 patterns per post. After 3-4 iterations — zero.

What genuinely works and what doesn't

Works:

  • Cringe Checker + Consistency Auditor together eliminate ~90% of AI-smell from final content
  • Content Repurposer saves me 2-3 hours per week — one article → LinkedIn + newsletter + Instagram caption + TikTok script automatically
  • Email Marketer writes onboarding sequences that make sense — because it has context from the CRO Specialist's full funnel view

Doesn't work:

  • Video Producer only works when you have recorded footage. It cannot generate video from nothing
  • TikTok Publisher is a placeholder — no TikTok API available for draft automation
  • Cold Outreach Specialist produces good templates but prospecting (the contact list itself) you must do manually
  • The Bridge (Layer 5) is only valuable if you are actively using team-dev in parallel — otherwise it is overhead

The critical rule: drafts only

Mar Team never publishes anything automatically. This is intentional and non-negotiable. Not because I fear mistakes — but because every post is my signature. No agent publishes anything in my name without my eyes on it.

Workflow: /team-mar → drafts land in the admin panel → I review and approve → manual publish or scheduled post after my sign-off.

Want to apply this?

If you want to build a similar content pipeline for your project — I can show you how to configure the agents for your brand and platforms. Check my services or book a consultation.