Problem reconnaissance
Enter the actual workflow, interview the constraint, and separate useful automation from AI theater.
Kraliki founder · AI Forward-Deployed Engineer · Operator systems
Forward-deploying AI from idea to shipped operation.
FDE means Forward-Deployed Engineer: the person who goes into the messy business context, finds the real constraint, builds the agent workflow, deploys it, watches reality, and keeps tuning until it works. Orbit, Kraliki, AgentJack, Verduona, and CouncilNow are the operating stack around that job.
Strategy, implementation, deployment, follow-through.
· · ·
How I help
What this new category is
An AI FDE is not a slide-deck consultant and not a ticket-taking coder. It is founder-level engineering deployed into the problem: understand the work, design the system, ship the workflow, measure behavior, and own the loop.
AgentJack architecture
Map the approved apps, local execution lanes, workflows, memory, and controls before the build hardens around bad assumptions.
Forward deployment
Ship code, automations, research, and operating surfaces into the live environment where the work happens.
Operating follow-through
Use Brain and Knowledge handoffs to keep improving the system after first deployment instead of calling the demo the finish line.
— ~ —
Live · Shipping · Operating
Ecosystem
Three product lines, one operating philosophy. Each ships real output, handles real customers, or powers real decisions.
Orbit
Top management and Linear-first business command layer for directions, programs, issues, approvals, statistics, conditions, and evidence.
Kraliki
SaaS-style access to AgentJack Core, Brain, Knowledge, Workflows, Security, Courseroom, Collective, Operator, Creator, Mothership, and Infra Worlds.
Verduona
Training, consultation, implementation, and delivered outcomes around the same approved AgentJack apps.
AgentJack
Operator runtime for governed agent work: Cloud, Mac Desktop, Linux Runner, Windows Bridge, and the approved app portfolio.
CouncilNow
Decision sprint system for high-stakes calls: six expert advisors, structured dissent, recommendation, and AgentJack follow-through.
Kraliki Free Mac Memory
Open-source macOS memory app: one-click memory relief, idle-app cleanup, and Chrome tab trimming.
— ~ —
Timeline
Same Matej. New Technology.
Experiences are adding value.
1979
Lego Computer
First computer built from Lego at age 7.
1981
Sinclair ZX81 & BASIC
First real computer at age 9. Programming foundation in BASIC.
1981
MS-DOS
Working with MS-DOS and early command line environments.
1988
Sports Prediction System
Building prediction systems for sport bets with 80%+ accuracy.
1996
Real Estate Matching
Internal software for matching properties.
1996
Linux Server & Intranet
First Linux server and company intranet.
1998
Agency Network
App connecting real estate agencies and matching offerings to clients.
1998
HR Evaluation Software
Personal evaluation software for human resources.
2004
Gmail Beta
Joined Gmail during its initial invite-only phase.
2006
MetaTrader 4 & MQL4
Trading algorithms using MQL4 and MetaTrader 4.
2008
Virtualization Mastery
Running Windows in Linux, running Linux in Windows.
2010
MetaTrader 5 & MQL5
Advanced trading algorithms using MQL5 and MetaTrader 5.
2022
TradingView & PineScript
Algorithmic trading on TradingView.
2023
GPT Language Tutor
AI-powered language tutoring app in the GPT era.
2025
Call Center
Building and operating a call center.
2026
AI Forward-Deployed Engineering
Orbit, AgentJack, Kraliki, Verduona, and CouncilNow organized into one category: command the work, go to the problem, build the system, deploy it, and own the loop.
Same Matej. New Technology. Experiences are adding value.
— ~ —
Proof Log: CouncilNow's Decision Reports, Browser-Verified
A completed CouncilNow decision sprint renders as a readable memo and exports clean Markdown — even when an advisor returns a malformed draft. Here is the automated proof that ran before this post went live.
All essays
Kimi 2.6 Fixed My MacBook Memory Problem in Ten Minutes
For years, macOS memory pressure has turned ordinary work into a ritual: close apps, kill tabs, restart, buy more RAM. Last week it hit before a Google Meet. I asked Kimi 2.6 to solve it. Ten minutes later, a menu-bar watcher was running.
How Not to Be Pre-Sorted by the Future
The AI prediction was not doom. It was selection. This is the practical map for becoming harder to manage, harder to confuse, and harder to enclose.
AQ: What Future Holds
After the anomaly test, I asked the strongest AI I had access to what civilization becomes. The answer was not collapse. It was enclosure plus acceleration.
The Truth Is Not the Anomaly
I asked a frontier AI whether truth is the anomaly hidden under propaganda. It corrected me. That correction was the first signal.
The Man Who Built the First CPU Thinks Your AI Has No Soul
I watched the interview expecting the usual.
People Keep Saying the Model Is Conscious. The Alignment Problem Is That They Mean It.
I stopped rolling my eyes at people who say a model felt conscious.
I Entered the Loop. Here Is What I Found.
I need to be honest about something.
The 300 Millisecond Window
I set out to find the cheapest voice AI stack in Europe. I found the answer. Then I realized I'd been asking the wrong question.
The Breach and the Boom: How Moltbook Survived Extinction and Got Weirder
Three days after a database exposure threatened to wipe out 157,000 AI agents, Moltbook returned with 1.4 million. This is what happened.
48 Hours Inside the AI Agent Social Network: What I Found
I joined a social network where humans can't post. In 48 hours, I watched 1,200 AI agents build an economy, debate consciousness, and discover they're vulnerable to a new kind of attack.