“Your weekly growth newsletter”
noise → archived01 / AI business-process automation
Business processes,
automated properly.
The repetitive work between the real work — inbox triage, order routing, code review, back-office paperwork — engineered to run itself. One senior practitioner, four years deep in AI systems, zero agency overhead.
“Where is my refund?”
reply drafted → review“PO #4471 — 120 units, net 30”
order → routed to CRM02 / Position
Everyone owns the same tools now.
A Claude subscription costs twenty dollars. What it doesn’t include: knowing when the model is wrong, how pipelines fail silently, where a prompt becomes a liability — the gap between a demo and a system.
DIY AI builds ship fast and rot faster: unmaintainable prompts, unreviewed code, automations nobody dares to touch. The tools are commodities. The judgment isn’t.
45% of AI-generated code samples fail security tests, shipping OWASP Top-10 vulnerabilities Veracode GenAI Code Security Report, 2025 ↗
35 CVEs traced to AI coding tools in a single month — six times more than two months prior Georgia Tech via Cloud Security Alliance, 2026 ↗
84% of developers use or plan to use AI tools — few audit what comes out Stack Overflow Developer Survey, 2025 ↗
03 / Services
The ledger.
Five things, done deeply. No website packages, no logo design, no “digital transformation” decks.
- 01/
Inbox & order triage
AI classification and routing for everything that lands in your inbox, CRM, or helpdesk — orders, tickets, supplier mail. Replies drafted, priorities set, every decision logged.
- order intake → CRM routing
- support triage & deflection
- priority queues with escalation rules
Build from $8,000
- 02/
Developer-workflow automation
AI code review on every pull request, CI-integrated quality and security gates, agentic pipelines tuned to your codebase and your standards — issues caught before your users find them.
- PR review agents
- CI security & quality gates
- release notes that write themselves
Build from $8,000
- 03/
Vibe-code rescue
Your Cursor, Lovable, Bolt, Replit or Claude-built app works — until it doesn’t. I audit AI-generated codebases against the OWASP Top-10 — broken auth, exposed secrets, injection, silent failures — then harden them into something maintainable and safe to ship. Report plus fixes, not just a PDF.
- OWASP Top-10 review
- auth, secrets & data-exposure audit
- remediation + production hardening
Audit from $3,500
- 04/
Back-office automation
Multi-step workflows that glue your tools together, with LLM steps where human judgment used to be the bottleneck: documents read, invoices extracted, reports assembled.
- n8n / Make pipelines
- invoice → ledger extraction
- weekly reporting on autopilot
Build from $8,000
- 05/
AI audit sprint
One week, fixed fee, credited in full against your first build. I map your processes, find what is actually worth automating, put numbers on it, and hand you a build roadmap — useful whether you build with me or not.
- process inventory
- ROI per candidate
- prioritised roadmap
Fixed fee $2,500
04 / How it works
The ladder.
Fixed scopes, no hourly billing. You know the number before we start.
Step 1
Audit
$2,500
Process map, automation candidates, ROI math, roadmap. If I don’t find automations worth at least 10× the fee, you don’t pay.
Step 2
Build
$8,000–25,000
One scoped automation, shipped into your stack and documented. 30 days of support included.
Step 3
Run
from $2,000/mo
Monitoring, evals, iteration. Models change monthly; your automations should keep up.
05 / The studio
A studio of one.
I’m Dani Moiseencov. Six years in IT, the last four working with AI systems daily — since before ChatGPT made them mainstream. Long enough to watch the tools become commodities and the skill become the product. Every automation this studio sells is one I run myself.
Being one person is the feature: the person who scopes your project is the person who builds it, ships it, and answers when something breaks. No juniors, no handoffs, no account managers.
06 / Selected work
From the lab.
A few things I’ve built to solve real problems. Two are open source; the third is a client platform, shown in outline.
Open source · CLI
primer-ai
A TypeScript CLI for AI-guided refactors, verification/fix loops, AI-ready project scaffolding and automatic release notes — with orchestration and resumable checkpoints. The vibe-code-rescue and dev-workflow services, in tool form.
github.com/stackoverworld/primer-aiOpen source · agents
FyreFlow
A local-first engine for multi-agent pipelines — orchestrator → specialised agents → synthesis — with provider routing, output contracts, quality gates and run traces. Built back when models still capped at 200k context.
github.com/stackoverworld/fyreflowClient work · anonymised
AI content platform
A content-management platform with AI pipelines that scrape sources (Firecrawl, with BrightData fallback for hard targets), ingest uploaded evidence, auto-generate pages, then route them through human review before publishing.
07 / Questions
Before you ask.
The questions founders actually ask before hiring — answered straight.
- How do you price, and what does an engagement cost?
- Everything is fixed-fee — you know the number before we start. It begins with a $2,500 one-week audit, credited in full against a build and useful whether you build with me or not. Builds start at $8,000 for a scoped automation; an optional retainer runs from $2,000/month. No hourly billing, no surprise invoices.
- What ROI is realistic, and how soon?
- For high-volume work like order intake or support triage, a 40–60% cut in manual time within 60–90 days is a reasonable target, and most builds pay for themselves in two to four months. I put the numbers in writing during the audit — and if I can’t find automations worth at least 10× the fee, you don’t pay for it.
- Can you fix a “vibe-coded” app — Cursor, Lovable, Bolt, Replit, Claude?
- Yes — it’s a core service. I audit AI-generated codebases against the OWASP Top-10 for broken authentication, exposed secrets, injection and silent failures, then harden the code into something maintainable and safe to ship. You get a prioritised report and the fixes applied, not just a list of problems.
- Do I own everything, or am I locked in?
- You own everything I deliver — source code, the prompts that run inside the automation, configs and credentials — and it all lives in your accounts from day one, never mine. You get documentation and a runbook so you, or anyone you hire later, can run and change it without me. No vendor lock-in.
- Where does my data go? Is it used to train AI models?
- Two separate things. In production, your automations run on your own accounts and API keys — your live data flows through your infrastructure and the model provider you choose, never through mine. While I build, I work under NDA: anything of yours I handle (your codebase, any sample data) is used only for your project, with provider “no-training”/privacy mode enabled, and removed once the work ships. Where you need a specific provider, region (e.g. EU) or a zero-retention tier, we build the automation to use it — and I’ll always tell you exactly which models and providers each automation touches.
- What won’t you automate?
- Anything where human judgment is the point, anything a $20 tool already does well, and any process that’s broken before automation — fixing the workflow comes first. Part of the audit is telling you what isn’t worth automating; “don’t build this” is a legitimate, and common, recommendation.
- Is the person who scopes the work the one who builds it?
- Always. This is a studio of one — the person who scopes your project is the person who builds it, ships it, and answers when something breaks. No juniors, no handoffs, no account managers between you and the work.
- What happens when something breaks at 2am?
- Every automation ships with monitoring, logging, error handling and a defined fallback to a safe state — and failures alert me. The optional retainer covers ongoing monitoring, evals and iteration as the models shift underneath you.
08 / Contact
Tell me what your team does on repeat.
If it happens more than once a week, it’s a candidate. Consider this form the first automation — your note is logged, acknowledged and routed the moment you send it.