Kamrun Nahar
wonderful and original
What people wrote about my work, in their own words, on the platform where they wrote it.
Kamrun Nahar
wonderful and original
Kyomuhendo Milliam
CTO, Asili Education
This reminds me of Dr. Tali Režun's comments during one of the 'Chasing Jarvis' sessions where he pointed out a major blind spot: while using AI to augment our workflows, we tend to assume the freed-up time is a break — and quietly fill it with endless scrolling instead of redirecting it toward production.
Reading your post felt like a cheat sheet for what to actually do with that time, Drazen Kapusta.
The coordination layer AI replaces — the chasing, translating, reporting — returns real hours. Dr. Tali Režun's point was that the productive response is to pour that margin into exactly what you describe: going deeper on your craft, teaching the person next to you, raising the level of the whole team
The extra hours are not a reward. They are raw material for becoming the player-coach.
Ever since that session I have been more deliberate with that margin. The player-coach is not built in moments of high visibility. It is built in the quiet investment between them.
What strikes me most is that I am getting these insights not in isolation, but layered on top of each other — Dr. Tali Režun in one session, this framework from you in another. Grateful for who I am slowly becoming because of the mentors at COTRUGLI Business School
Ana Romac
Sales Director, Iskon
Last week I was in Belgrade for the first time - incredibly beautiful and alive everywhere!
Last week I really studied again - awake until 4, 9 classrooms.
Last week I went out more than I have in the whole year so far (I need to work on that, I see it's going quite well everywhere...)
Last week I built my own application and, for the first time outside of everyday life, touched the real possibilities of AI..
When I thought about enrolling in an MBA program, I wondered what I can actually get today with my experience and knowledge at my fingertips. And all the conversations always ended in - networking. I just didn't realise that by that I meant "living together" for 5 days and nights. That brings another dimension, quite different from a few well-organised meetings. That brings memories for a whole life.
Thanks to professors Dr. Tali Režun and Dr. Nicolas Kfuri for two wonderful and inspiring modules.
Thanks to my "networking" for all the interactions, be they over coffee, in markdown files or in Asia. I won't tag you, you know who you are and how it was for us. Some memories stay where they happened (and in any case - in the WA groups).
I'll just say - I'm still smiling and can hardly wait for ZG
Translated from Croatian
Prošli tjedan sam bila prvi put u Beogradu - nevjerojatno lijep i posvuda živ!
Prošli tjedan sam ponovno stvarno studirala - budna do 4, 9 učionica..
Prošli tjedan sam izišla više puta van nego u cijeloj godini dosad (moram na tomu poraditi, vidim da mi ide skroz ok...)
Prošli tjedan sam napravila vlastitu aplikaciju i po prvi put izvan svakodnevnog života dotaknula stvarne mogućnost AI-a..
Kad sam razmišljala o upisivanju MBA programa, vagala sam što danas sa svojim iskustvom i znanjem dostupnim na dlanu zapravo mogu dobiti. I svi razgovori su uvijek završavali na - networking. Samo što pod tim nisam podrazumijevala da "živimo zajedno" 5 dana i noći. To nosi jednu sasvim drugu dimenziju od nekoliko odrađenih sastanaka. To nosi uspomene za cijeli život.
Hvala profesorima Dr. Tali Režun i Dr. Nicolas Kfuri na dva predivna i inspirativna modula.
Hvala mom "networkingu" na svima interakcijama, bile one na šanku, u markdown fileovima ili u Aziji. Neću vas tagirati, znate vi tko ste i kako nam je bilo. Neke uspomene ostaju tamo gdje su se dogodile (i za svaki slučaj - u wa grupama).
Samo ću vam reći - još uvijek se smijem i jedva čekam ZG
Snežana Ilić
Business Development Manager, COTRUGLI Business School
Thanks to Professor Dr. Tali Režun for his vision and leadership, and to all participants for their enthusiasm, openness, and engagement throughout the program.
What stands out most is not the technology itself, but how quickly meaningful ideas can become real solutions when expertise meets the right AI tools.
The projects showcased during #ChasingJarvis are a powerful reminder that the conversation is no longer about whether AI can create value. The real question is how organizations and professionals can leverage it to solve complex challenges, move faster, and unlock new possibilities.
Events like these remind us that while technology is advancing at an extraordinary pace, it is still people—their knowledge, creativity, and collaboration—who make the real difference.
Exciting to see this happening in Belgrade and inspiring to witness so many professionals turning vision into execution in just two days.
[Second post]
Excited to see participants moving from ideas to working MVPs in just a few days.
Proud to be part of the COTRUGLI team and to support learning experiences like this.
Matthys van Rooyen
Machine Learning and Software Engineer
This week in another session of Chasing Jarvis with Dr. Tali Režun, the focus was a practical one: setting up your own personal (or organisational) AI agent.
We were introduced to OpenCode, which Dr. Tali rates for being open source, giving access to a range of free models, and letting you plug in local models too.
The setup itself is simple. Using Dr. Tali's framework, you hand the base coding agent one predefined AGENTS.md file and let it build itself. It even interviews you about your goals and details to personalise the agent, then self-evolves from there.
We also covered how you can easily install MCPs, isolated to each agent, by just asking the agent: an Excel MCP and a mailing MCP, plus The Curator, Dr. Tali's excellent "second brain" for long-term context, connecting your agent to your own knowledge.
The final step: taking the agent autonomous with CRON jobs, so it runs on a schedule without you.
From one markdown file to an autonomous digital worker. Great session, and a strong toolkit to keep building on.
COTRUGLI Business School
Godfrey Muhando
AI & Digital Expert
Yesterday's session in Chasing Jarvis 5 with Dr. Tali Režun was another exciting step into the evolving world of AI agents and autonomous software development. We were introduced to OpenCode AI, a powerful AI coding agent, and explored everything from installation and integration to navigating the terminal, setting up project memory, creating .md files, and using effective prompts to maximize the agent's capabilities.
One feature that immediately stood out to me is its flexibility. OpenCode AI works with multiple AI models, making it easy to integrate into existing development workflows. I was also impressed by its terminal-first approach, project memory, and support for multiple AI agents working on different tasks simultaneously.
As someone exploring AI-powered development, I'm excited to continue learning how this tool can help transform ideas into real-world solutions faster and more efficiently. I look forward to applying these concepts in my upcoming projects throughout my MBA journey at COTRUGLI Business School and beyond, creating practical business solutions with the power of AI.
A big thank you to Dr. Tali Režun and Drazen Kapusta for the insightful session and to my classmates for the engaging discussions. Looking forward to what's next in this incredible learning journey!
Priscile Samou
Operations & Supply Chain Professional
Thank you, Dr. Tali, for another outstanding session. Your hands-on teaching approach makes complex AI concepts practical and accessible. Truly inspiring. Looking forward to applying what we've built today!
Petar Urdešić
Banking Operations Executive
Dr. Tali Režun thank you for sharing your knowledge and for the inspiring 2 days - it was very useful!
Marko Hodak
Freight, Logistics & Supply Chain Advisor, Bugatti Rimac
Thank you, professor Dr. Tali Režun!
It was a real pleasure, you are truly a man of great knowledge and an excellent presenter. I am looking forward to continuing my learning path with you as there are a lot possibilities for upgrading my knowledge and creating high value both for myself, and our work organization.
Let's shape our tomorrow by designing it together, today!
Dejan Romih
Founder & CEO at MountVacation
thank you Tali - I am looking forward for a big transformation in online travel.
Enrique Barclay
Inside Sales Engineer @ Commvault | Ex-Founder, Digital...
The honest part is what makes this credible: the dark green Saturdays were debugging, not heroism, and agents deploying to the wrong project is the real texture of building this way. Your orchestration point is the whole thing. Knowing when to checkpoint, when context is rotting, when to kill a session, that is judgment, and it never shows up in a commit graph. I have felt the same shift: the code stopped being the constraint, and deciding what is worth building (then verifying it is actually real) became the job. How do you catch a session that has gone bad before it costs you, a smell test, or hard signals?
Enrique Barclay
Inside Sales Engineer @ Commvault | Ex-...
The 80% mark is the exact line I watch too. What made it click for me was treating the handoff md as a real artifact, not a scratchpad: I write it like a new engineer is taking over cold, then let the session die. Do you keep one rolling spec per project, or branch it per major feature?
Abhishek Srivastava
Founder, Eleven11 | Architect: your AI agents do the work, your team ...
Dr. Tali Režun The line that got me is "a short list of what only I can do." I run an orchestrator with a roster of hired agents and that list became its own column on the home page, Waiting On You. The whole thing is a machine for keeping that column short and honest.
Marius Jost
SEO Services | ex-Burson-Marsteller | Connecting ...
OK, this opened my eyes. We will try this here as well now. I have heard from this approach and it does make sense. It had always been obvious to me that for certain tasks, the flagships are overkill (and you overpay). Thanks!!
Sarah Wahinya
Web3 Ecosystem Strategist | Developer Relations & Partnerships | ...
The build & debug phase (2)is where things gets real Prof Tali, because this is the chaos management, one juggles between debug loops that drain time and mental energy, Agents that gets 70% and miss critical edge cases and one end up trying to decide to trust the agent or step in manually 💯
Nardus du Plooy
Vanguard MBA
Three weeks into Chasing Jarvis, attempting to build my first MVP as a complete non-technical Vanguard member, this article stopped me in my tracks.
I'm already living the cycle you describe:iteration → failure → discovery → refinement - often within the same day.… more
Nardus du Plooy
Cotrugli Business School Vanguard member
I installed the Curator and then ingested Drazen Kapusta's Art of Vanguard Leadership. What came back wasn't a summary. It was 20 structured wiki pages: NEO Era, Vanguard Leadership, Wolf-Dog-Sheep Patterns, Fear as Data, Reputational Capital. Each one was cross-linked to the others. I then pulled those pages into the Navigator app I've been building for the Vanguard MBA, as a Second Brain Vault the students can query on their phones. The distinction between storing and compounding became real in that moment. Storing is a pile. Compounding is architecture. I then asked The Curator "What is the Sheepdog principle?" and it answered from the book with a source citation. That's not search. That's thinking.
VTI
Company page
Only 9 MCPs? That's practically a full dev team in disguise, Tali! It's impressive how far this approach can go before hitting the complexity ceiling.
Ahmed Abulkhair
AI Scientist @ Incept Labs | Machine Learning & LLM ...
The way the agent builds its own folder structure from that single identity file is a sharp way to keep things simple and avoid config chaos. Also, the step where it interviews you one question at a time about your role feels like the kind of onboarding that actually works, no fluff.
IndaPoint Technologies Pvt. Ltd
Company page
We think this is a strong step toward making knowledge management more accessible without sacrificing ownership or portability. Keeping the knowledge base as plain Markdown on the user's own machine is especially valuable it reduces vendor lock-in while making the information usable across tools, agents, and workflows.
The agent memory layer is also interesting because persistent context can reduce repeated discovery and help agents build on previous decisions rather than starting from scratch.
How do you see teams balancing shared organizational knowledge with privacy and individual control as agent memory becomes more common?
Raymond NISINGIZWE
CEO& FOUNDER: Digital GreenShield Africa an...
The Curator brain is very valuable for our Digital GreenShield Africa because it transforms scattered agricultural data (farmers, climate, crops, finance) into a connected knowledge system. Instead of storing information in isolated dashboard, DGSA can use a knowledge graph where every farmer, weather event, and crop is linked, farmer ,NGOs ,Government to interact ,making it easier to detect hidden patterns like drought → low yield → loan risk. This helps DGSA move from simple data reporting to intelligent decision-making, prediction, and real-time insights for farmers and financial partners.
Nardus du Plooy
Cotrugli Business School Vanguard member
The ingestion bottleneck is real. It's not a lack of curiosity, it's the friction between saving something and actually making it yours. I think the domain-based approach is the smartest part. One focused wiki per topic, getting richer with every source you add. That's what makes this different from every "save everything" tool that… more
Valentine Nguemne
Software Engineer, Head of engineering, ...
Just cloned and tested it : amazing!
Though i exagerated and exhausted my free quota for gemini already.
Can't wait to see it grow into using a free LLM hahaha.
Nardus du Plooy
Cotrugli Business School Vanguard member
Really insightful and thought-provoking
I recently had to unpack my municipal utility charges against my lease agreement — something that would normally have taken an entire evening of cross-referencing invoices, clauses and billing periods. With AI support, I had clarity within minutes.
I didn't use a multi-agent setup myself, but your explanation of that architecture is fascinating. It highlights how critical context engineering is — and how structured reasoning tools are steadily lowering the barrier to navigating complex systems like law, contracts and public billing.
Raymond NISINGIZWE
CEO& FOUNDER: Digital GreenShield Africa and ...
Great insights. This is especially relevant across Africa where many entrepreneurs and small businesses cannot easily access legal services. If AI can help people better understand contracts, regulations, and risks, it could empower thousands of businesses to operate more confidently and sustainably.
Kaouther ZAAFOURI
CEO and Founder at Zaafouri International Consulting - Environme...
Congratulations to Dr. Tali Režun and Anton Dobrina for this insightful contribution. Fascinating perspective. If law can be understood as code, AI has the potential to decode complex systems and democratize access to expertise that was historically limited to a few. This approach could extend far… more
Ahmed Abulkhair
AI Scientist @ Incept Labs | Machine Learning & LLM ...
That moment when a real customer hits an edge case you forgot about feels both frustrating and like pure validation. Fixing that silent 2, 000-character limit and then quadrupling it shows how real use cases force you to build for reality, not just demos.
Trino Gomez Gamboa
Business Leader & CEO & Co-Founder | Investment & Business ...
A strong example of how AI can move beyond simple chat into a practical, compliant business tool that improves both customer experience and internal insights.
Dejan Romih
Founder & CEO at MountVacation
Thank you, Dr. Tali Režun, for a great day and inspirational new knowledge on AI and the future of coding.
Drish Mali
ML Engineer
Dr. Tali Režun Great insights! My 'AHA' moment was finally understanding that Phase 1 is the most important step. I used to spend all my energy on the build (Phase 2), but your framework helped me realize that rich, structured context is what actually allows agents to perform. Moving from a developer mindset to a product architect mindset changes everything. Thanks for sharing the playbook!
Carlos T.
Chef Projet, Msc (Santé publique en Controle des ...
For me, as a young entrepreneur/student and a novice in the use of AI, reading your articles has made me realise very quickly that AI is no longer a simple tool to be used as a mere 'shortcut'. It is a 'shortcut' system that I must fully master; I must integrate it; it is a guiding thread to boost my efforts and always go further and higher. To this end, I must research and adapt my documentation to the aim, objectives, and expected results, and above all, always remember to update my documentation.
For example, as a biologist and public health specialist, AI needs to be intelligently integrated into my daily workflow... Quality of results; client satisfaction; expanding my client base; selecting my community health studies; analysis, monitoring, and evaluation, etc....
Written in French
Johan Landman
CSGI
The articles reinforced for me that the real challenge with AI isn't technical adoption, but leadership readiness. As AI becomes embedded in day-to-day operations, leaders must understand the AI environment — its constructs, functionality, and solution boundaries — well enough to set the right guidance,… more
Carlos T.
Chef Projet, Msc (Santé publique en Controle des ...
Although I am no expert in all things IT and AI, I have read the various articles very carefully. My thanks to Dr TALI, as these articles are far from being mere academic papers; they also serve as a guide and a catalyst for our successful experiences with AI, and even for our activities unrelated to AI.
The driving force behind everything we do remains: our documentation. What do we want to achieve? Where will we find our information? When and how will we put it to good use? And above all, why use one form of documentation rather than another? Our documentation remains our cornerstone, for if it is used effectively and tailored appropriately, it will pave the way and facilitate the rest of our work. And the same applies to the use of everything surrounding AI to improve our daily lives.
Nouman Nawaz
Helping AI startups build investor-ready MVPs and naviga...
The 6-senior-devs estimate is the fork in the road for most non-technical founders. A few outsource it, a few kill the idea, and most end up with something built that was never quite what they meant.
What strikes me in your case study is the decision layer: 300+ orchestrator sessions, and you were the one deciding what should exist, not what should compile. That's the part nobody talks about. The orchestration tooling is getting good. The bottleneck is still the person who can hold the product vision clearly enough for 400 agents to stay aligned for 8 weeks.
Working with non-technical founders on their AI builds, that's where we spend the most time. Not on the models. On the brief.
Mislav Bašić
CEO at 4uHa
Dr. Tali Režun I've had some really inspiring chats this year with Drazen Kapusta and you. They helped me shape my goals for the year ahead. Cheers, and looking forward to seeing you again soon!
Sarah Wahinya
Web3 Ecosystem Strategist | Developer Relations & Partnerships | ...
This really resonates. I've hit that 70–80% wall with single agent workflows, especially as complexity grows. The shift toward orchestration and treating the spec as the real source of truth feels like the actual unlock. Still exploring how to structure that 'score' well enough to support that level of coordination
Marko Hodak
Freight, Logistics & Supply Chain Advisor | 15+ Years at DHL, ...
The level you reached and the amount of details on your hub is "out of this world". You are like a living database. There is nothing left to say, just to admire your work, and try to learn as much as possible from you. 👏
Nardus du Plooy
Cotrugli Business School Vanguard member
I agree. The bottleneck is no longer intelligence. It's trust, permissions, and coordination. I'm seeing this directly while building the Cotrugli Navigator MVP. The AI capability is moving unbelievably fast, but the moment you move from demo to real environment (or close to), the walls appear immediately: siloed data, API restrictions, governance, compliance, and human hesitation. The hardest orchestration problem is no longer the agents. It's the humans around them. And keeping up with the shifting tools and infrastructure is a full-time job on its own. Quite overwhelming. My biggest immediate wall is the social API problem. The Engage tab requires the student's own opinion first, then drafts through three iterations grounded in their thinking, but they still copy and paste manually at the end because the gate is closed. Is structured agent access to LinkedIn something you see opening in the next 12 months, or do we build assuming the wall stays?
Raymond NISINGIZWE
CEO& FOUNDER: Digital GreenShield Africa and Engine...
This really changed how I think about AI tools.
It's not about which tool is best—it's about how well you guide it. The same tool can give very different results depending on the context you provide.
Also realizing that cost and performance reflect how much… more
Anena Sharon
Faith Driven Entrepreneur | Business & Leadership Coac...
This article perfectly captures where I am right now, aware that something big is happening with AI, but unsure how to properly go about it. The idea of investing time in research and planning before touching any tools is a mindset shift I had to embrace from the very beginning of my EMBA. Thank you for mapping the… more
S M Faidarul Alam
Ex. Hotelier | Tourism and Hospitality Management | ...
This shift from solo instrument to symphony captures it perfectly. The structural gap between orchestration-led development and traditional teams isn't about skill—it's about rhythm. One moves at the speed of context, the other at the speed of meetings. That collision is going to define the next few years.… more
Augustine Mutua
IT Consultant | Tech Solutions Expert | Vanguard MBA
Orchestration is not magic — it depends on strong context. Phase One is everything: research, context engineering, and clear documentation are what make the speed of Phase Two possible. If the input is vague, the outcome will be vague too.
It also made me realise that the advantage is shifting toward… more
Jannes du Plooy
Founder, AI Notary: making every AI decision and every ...
I can totally relate. A new breed of "developer" has entered the arena, bringing a completely different perspective to the party. The approach (i have learned through several small simple projects) described by you is absolutely crucial. No shortcuts.
Alvin K.
Operations Manager at Nikola Energy
The metaphor of the symphony vs. the solo instrument is spot on. Context collapse at that 70-80% mark has been the 'glass ceiling' for so many AI projects. Seeing how parallel worker and review agents solve for that scale is a game-changer for complex builds.
Johan Landman
CSGI
One of my primary challenges is determining where to begin and managing the uncertainty surrounding what lies ahead. At times, this uncertainty leads to overthinking, which can hinder progress before a task has even started. This article offers valuable clarity on how to approach the initial stages and outlines expectations… more
Mário Alexandre | AI Workflows
The separate context per worker is the underrated part. One trap: a worker with no model set inherits the orchestrator's model, so simple ones to Sonnet only holds if each subagent pins it. Writing the split down as a plan is what kept it from drifting for me.
Valtteri Melkko
This sounds amazing - well done. I think it's important for the ecosystem of "harness-agnostic tools" and "cross-harness tools" to develop - been experimenting in that space quite a lot myself as well. Going to check out AgentLink, it sounds super-interesting.
Mehdi Zare
This is the part that feels durable: a shared handoff layer matters more than which model opens the session. The context contract is the real harness.
aditya bhatia
Framing the context window as finite short-term memory is a helpful lens. It also explains why long sessions drift: older instructions get crowded out by recent tool output well before the window is technically full.
mao mao
shared memory is the bit my agents keep missing; I've had to paste the same project constraints into three separate sessions before.
Pablo
menu bar at-a-glance for a context engine is such a Mac shape. love that the wiki surface lives there.same instinct on my side with Ventus. keep the controls in the menu bar and stay quiet until the machine spikes.
Harry Ashton
Oh incredible I did think of Obsidian as you read this and saw you're using the map for it. Very useful indeed, a common across all teams/projects but I wonder how efficient this is as opposed to maintaining skills? Or is that a work in progress? :)
Ademola
Yeah, that was the part that clicked for me too
The more I dug into it, the less it felt like a tool for "AI memory" and more like a layer that lets the context survive while everything around it keeps changing
I am glad the visual got that across 🙏
Ademola
“Your context survives” is the promise
“Here is exactly what the agent is starting with” makes it to be tangible
That feels like a really strong visual story too
Nizm@x
two agents planning my ski trip while i just show up is the dream.. the mv travel mcp server looks interesting ser, lets test it out
suna | building in public
The part that changed the most for me: who I am and what the business needs, in one file Claude reads first, every session.
Long time before that, I re-explained my business every morning. Now the first message already knows. 🙂
Sushant Shinde
Co-Founder, Enerscript · CPO, CoreLogiQ | Product ... (3rd+ connection)
The shift to context-driven MVPs with AI agents democratizes product building. It genuinely empowers business professionals to ship solutions faster.
Johan Landman
CSGI
Dr. Tali Režun Our “Chasing Jarvis” MBA journey really reinforced this for me. Understanding AI agents is one thing, but actually building and experimenting with them changes the perspective completely. Practical AI knowledge like this will be an increasingly important skill for future leaders. Great initiative💪👌👍
mayfair-workroom
Runs https://mayfair-workroom.com/ (interiors business, per own comment)
Really enjoyed this. The process feels clear, practical, and honest about where AI coding helps and where the hard work still stays manual. I liked the focus on context, structure, and discipline before speed. I run https://mayfair-workroom.com/ , so my work is in interiors, not software, but the same idea applies. Good results come from a strong foundation, clear planning, and careful execution. This was a smart read.
S M Faidarul Alam
This is an exceptionally well-documented, battle-hardened methodology. What sets this piece apart is the maturity of perspective: the clear delineation between being a "junior developer using AI" versus becoming a "product architect directing AI teams." The three-phase framework—Context Engineering, Build, and Production—provides a much-needed antidote to the "vibe coding" chaos that often dominates discussions about AI-assisted development.
The emphasis on Phase One as the non-negotiable foundation is the most valuable insight. Framing documentation not as overhead but as the primary driver of agent efficiency reframes the developer's role from implementer to system designer. I particularly appreciate the honest accounting of the debugging phase; acknowledging that agentic testing still carries significant friction (token costs, login walls) adds credibility that a purely promotional take would lack. This is a practical, real-world guide for anyone serious about scaling their output without sacrificing production-grade discipline.
Ehsan Yazdanparast
The image of two capable agents and one person carrying messages between them is uncomfortably accurate. I have done that same copy and paste loop between tools, and it feels productive until you notice a whole evening has disappeared. I also liked that you spent years on the question of who owns a message before you tried to make the setup easy for someone who never opens a terminal. That order is rare, and it usually shows later in the quality of the tool.
Tejas Sharma
been building something like this too. the graph view is where it gets interesting. around 30 to 40 sources in, you start seeing clusters you didn't plan for. connections between things you read months apart that you never consciously linked.
Robin Good
Wow. This is really very good.
Thanks for having created The Curator and having made available to everyone!
.
I am trying to buy some extra time to test and explore it as much as possible.
A few small feedback points from a non-expert, non-technical user:
.
Lious
The explicit separation between technical maturity and the unproven real-group test is the strongest part of this account. A shared knowledge system can pass concurrency and deletion tests yet still fail socially: unclear ownership, unresolved contradictions, or corrections that never propagate. For the first cohort, which evidence would matter most beyond uptime: correction latency, source attribution, conflict resolution, or whether contributors can explain how a shared conclusion changed?
NO TRACKING COOKIES
This site sets no cookies while you read it, and its analytics count pages, not people. Lumina, the assistant on every page, runs in the EU and is built for the GDPR and the EU AI Act.