live businesses · real automation

AI automation for websites and operations.

I build agent systems for real businesses: websites that publish, booking flows that qualify demand, music tools that analyze tracks, support workflows that answer customers, and e-commerce operations that monitor products, SEO and competitors. The proof is live across Optagonen, Mixanalytic and Cigge — and the terminal on this page runs against this site's real public MCP. Start with one workflow.

uygarduzgun.com — public MCP

scope: public.read

visitor$

3companies in production

27tools behind auth

500+pages by workflow

10locales automated

4public read-only tools

Three live businesses. Three different kinds of automation.

This is not one demo wrapped in a landing page. These are active systems in education, music tech and e-commerce.

Optagonen

Agent-assisted workflows run key parts of the operation: website, content, bookings and workshop logistics for creative school workshops. Public proof: 500+ workshops since 2015, 150+ municipalities, 2,000+ students.

optagonen.se

Mixanalytic

AI music analysis product in production — upload flow, 21 analysis modules, customer support and algorithm feedback loops. Public proof: 34,000+ tracks analyzed, 5,400+ registered users, 4.8★ from 143 reviews.

mixanalytic.com

Cigge

Agent-first e-commerce operation covering product and content workflows, SEO, repetitive admin and competitor monitoring around a live storefront. Details are intentionally limited because this is an active retail operation.

www.cigge.se

// live architecture

This is the pipeline that writes this site.

Twelve agents, real feedback loops, one draft-first rule: nothing publishes without a review point. Click any agent to see its job — this is the same flow I run from my admin panel.

→ review← revise

SC Intelligence

3 keyword opportunities found · avg position 12.4

// under the hood

The pipeline is only the top layer.

Underneath sits a full agent infrastructure: an MCP server architecture with machine-readable discovery, a public read-only endpoint, and write-capable tools separated behind their own keys. Every workflow on this site — content, SEO, translation, publishing — runs through these layers.

How the MCP architecture works
  1. Agent card

    public

    Machine-readable discovery — AI agents find out what this site can do without asking.

  2. API catalog · llms.txt

    public

    OpenAPI metadata, markdown rendering and docs built for crawlers and agents.

  3. Public MCP

    public

    The read-only endpoint the terminal above is talking to. 4 tools, no credentials.

  4. Customer API

    🔒 key required

    Scoped pilot workflows — one workflow per private key, rate-limited, draft-only.

  5. Owner MCP

    🔒 owner key

    27 write-capable tools: content pipeline, SEO, translation to 9 languages, images, publishing.

  6. Admin API

    🔒 admin only

    Direct blog, settings and pipeline operations. Never exposed to agents or customers.

// no forms, no calendly maze

Skip the contact form. Brief my agent instead.

Three questions. Leave your email at the end and the brief goes straight to Uygar. A human answers, usually within a day. Scoping calls are free.

This agent is deliberately scripted — the live AI runs behind auth, where it belongs. That boundary is the whole pitch.

UA

Uygar's booking agent

online — scripted, zero LLM cost

// faq

The short version

You don't need a big AI project. You need one useful workflow that stays under control.

Read the technical write-up
Do I need to be an enterprise company?

No. A good first project can be as small as a website that needs SEO drafts, page updates, lead follow-up, support replies or product text. Enterprise teams usually need more review, logging and integration work.

What can a normal website owner automate first?

Start with tasks around the website: turning notes into drafts, improving existing pages, writing FAQ sections, finding internal links, summarising enquiries, preparing newsletter drafts or creating SEO briefs.

Is anything write-capable open to the public?

No. Public discovery is read-only. Any workflow that creates, updates, publishes, translates, generates images or spends AI budget requires a private key.

Who is already running these patterns?

Uygar is — he's the AI automation consultant behind all three businesses on this page. Optagonen uses agent-assisted workflows around content, bookings and workshop operations. Mixanalytic uses AI systems for music analysis, customer support and algorithm feedback. Cigge is an agent-first e-commerce operation where many repetitive product, SEO and competitor workflows are automated, with internal details intentionally limited.

What does a pilot cost?

Pilots and advisory engagements are fixed-price after scoping. The price depends on the workflow, integration depth and review needs. Discovery calls are free.

How long does a pilot take?

Most pilots ship in 2 to 4 weeks. Scoping and key issuance in week 1, first reviewable output in week 2, and iteration or handover in weeks 3–4.

What does a first pilot usually include?

Usually one content, SEO, website, e-commerce or internal operations workflow. It must have a clear input, useful output and a review point before it can expand.

Can technical teams review the architecture?

Yes. There is a technical proof article that explains the discovery layer, MCP split, auth model and private workflow boundaries. Code-level questions are welcome before signing.

// your move

Want part of your business to run like this page?

Send the website, the bottleneck or the architecture question. I'll tell you if it should be a small automation, a scoped build — or left alone.

Based in Gothenburg, Sweden. Working with European and global teams.