TL;DR

In my last post on Composable Task Execution Graphs (CTEGs), I introduced a modular structure for agentic execution. But execution comes after planning.

This follow-up introduces the Unfolding Framework, a concept engine built atop the Five-Fold Meta-Pattern, a recursive cognitive structure by NerdyRodent. The Unfolding Framework helps transform vague, complex business goals into constraint-aware, impact-prioritized execution graphs.

The system goes from:

Intent → Interpretation → CTEG → Execution → Feedback → Iteration

The Planning Problem

When someone says “launch our product in six weeks, we’ve got $3k and two freelancers,” they’re not assigning tasks — they’re initiating a mission.

Most AI systems today respond to that by generating a to-do list. But achieving a desirable outcome is more than enumeration. It’s about navigating ambiguity, surfacing constraints, balancing tradeoffs, and identifying what’s worth doing first.

CTEGs give us the container to execute tasks. But what fills the container — and how we decide that — is a different class of problem. It requires opinionated selections.

The five-fold meta-pattern provides a way to unfold the mission into an iterative discovery and decision process.

The Need to Think Before Graphing

I tried prompt chains, templates, playbooks. They worked — until the use case changed, or the context shifted, or a new constraint surfaced mid-execution. Even with a CTEG engine in place, I kept facing the same bottleneck: how to reason from a high-level mission to a viable set of tasks, without relying on static trees.

That’s when I found the Unfolding Pattern of Being, a compact and deeply recursive cognitive framework by NerdyRodent. It offers five semantic lenses — Order, Chaos, Process, Shell, and Recursion — that work together to analyze complexity and guide unfolding systems.

I adapted the Unfolding Framework as the planning engine — a “framework application protocol” for generating CTEGs from high level business goals.

How the Unfolding Framework Works

The process begins with intent — intent with some context, usually insufficient context. The unfolding framework breaks that intent down through five lenses:

• ORDER: What is known, stable, or required? What’s already fixed?
• CHAOS: What’s ambiguous, missing, or in flux? What needs discovery?
• PROCESS: What transformations are required? How should flow emerge?
• SHELL: What constraints define the feasible space? Budget, time, capacity?
• RECURSION: Where do we need feedback loops? Where should plans adapt?

Each of these informs the others. The more intent unfolds, the more the picture clarifies. Crucially, a ranking logic is needed to prioritize — the system applies the ICE scoring system:

ICE = (Impact × Confidence) ÷ Effort

For every option surfaced — whether it’s a discovery task like “interview beta users,” or a build task like “create teaser page” — the system applies the ICE lens to prioritize what makes it into the CTEG.

High-impact ideas with uncertain payoff get broken down further or pushed to conditional branches. Low-confidence but high-impact options may need lightweight exploration tasks to validate before inclusion. The CTEG evolves as the system makes decisions using this scoring rank.

This transforms prioritization from list-making into strategic unfolding.

A Real Example: Planning a GTM Campaign

Say the prompt is:

“We’re launching our fintech product in 6 weeks with $3k budget and two freelancers. What’s the most effective go-to-market plan?”

The AI engine doesn’t jump to a content calendar. It starts unfolding:

ORDER tells the system the goal is sign-ups, launch awareness, and perhaps pre-conversion.

SHELL limits it: no paid ads, no designer, limited hands.

CHAOS shows me the target persona isn’t yet clear. Do we have messaging? A unique value prop? Competitive context?

The system inserts tasks to surface these gaps — small loops to reduce CHAOS.

PROCESS helps the system decide: Should content come before persona validation? Can outreach and landing page build happen in parallel?

Then comes RECURSION. Nothing is final. If early teaser traction underperforms, the system adjusts messaging. If email CTR is low, it reroutes into message testing loops.

At every fork, ICE scores guide decisions. “Email funnel A/B test” might rank high (low effort, medium confidence, medium impact). “Hiring external motion designer” scores low (high effort, low feasibility under SHELL).

The result? A lean, high-leverage, feedback-resilient CTEG — not written ahead of time, but unfolded through structured reasoning.

Reflection

CTEGs give the system the how of execution. The Unfolding Framework gives it the why of structure.

Rather than hardcoding plans or blindly generating task trees, I’ve built this AI to use the Five-Fold Meta-Pattern to surface what matters, when to act, and how to adapt.

This isn’t about replacing current agent systems — it’s about empowering them to think like strategists before they act like assistants.

It’s planning not as specification, but as recursive interpretation.

And in that sense, task planning really is the new programming — except now we’re writing in the grammar of intent.

Takeaways

• Use the Unfolding Framework to generate intelligent, adaptive CTEGs
• Apply the ICE scores (Impact × Confidence ÷ Effort) to prioritize task paths
• Missions aren’t AI-generated — they’re unfolded by AI with the meta-pattern framework

About the Author

PAK is a hands-on engineering leader with a track record of building products and teams across Web3, AI, and gaming startups. Known for his obsessive attention to structure, he thrives in chaotic environments by applying systems thinking, frameworks, and operational playbooks to drive clarity and speed.

A firm believer in human-AI collaboration, PAK sees generative AI not as a replacement for talent, but as a force multiplier — freeing builders and thinkers to focus on what truly matters.

🔧 Explore the AI tool Pak is building — [join the waitlist here].

Task Planning Is the New Programming was originally published in Technology Playbook on Medium, where people are continuing the conversation by highlighting and responding to this story.

I. What Is a CTEG?

A Composable Task Execution Graph (CTEG) is a task orchestration system structured as a directed acyclic graph (DAG) that models the execution plan of a project.

Each node represents a task and is classified into one of four categories:

1. AI-executable
2. Internally executable
3. Outsourced
4. High-complexity or unknown (requires decomposition)

Edges encode dependencies and weights represent cost, time, and risk score.

Any node that requires decomposition triggers a recursive expansion into a sub-CTEG. This is a classic divide-and-conquer problem-solving, generating a hierarchical tree of nested DAGs. The end result is a collection of “mission” graphs that complete a project through modular task breakdown and execution.

II. Thinking Models: System 1 / System 2

• System 1 (fast, automatic): Tasks executable by LLMs, agents, or deterministic logic
• System 2 (slow, deliberate): Complex tasks, uncertainty, or decisions needing exploration

CTEG bridges both, eliminating the need for manual discovery of automation workflows or ad hoc constraint-solving. It enables efficient execution of automated tasks while simultaneously navigating complex decision paths in an observable and optimizable graph.

This also aligns with the multi-shot reasoning paradigm in LLM usage:

• One-shot frameworks rely on a single, do-it-all prompt in one pass (e.g. prompt-in, answer-out)
• CTEGs are inherently multi-shot, chaining reasoning and execution over recursive decompositions, memory and feedback.

III. Architecture Overview

Core Modules:

• DAG Engine: Stores and traverses nested task graphs
• DEA Router (“MoE-like”): Routes tasks to domain expert agents
• SOP Layer: Retrieves/generates SOPs per task
• Resource Graph: Tracks internal/external resource availability and cost
• Optimization Layer: Computes least-cost paths under constraints
• Execution Engine: Manages agent workflows and execution tracing
• Feedback Loop: Updates edge weights, SOPs, and DEA models via performance telemetry

Data Structure:

TaskNode = {
  "id": "task-uuid",
  "title": "Design Homepage",
  "type": "internal" | "ai" | "external" | "high_complexity",
  "status": "new" | "in_progress" | "complete",
  "dependencies": ["task-id-1", "task-id-2"],
  "weight": {
    "time": 6,
    "cost": 1200,
    "risk": 0.4
  },
  "sop_ids": ["sop-uuid", "sop-uuid-2"]
  "agent": "dea-marketing",
  "subgraph": "nested_cteg_id"
}

Graphs can be stored as collections of TaskNodes + dependency maps.

IV. Implications

CTEGs are not just task engines — they are adaptive cognition graphs for AI-human systems. They reduce execution ambiguity, enable explainable orchestration, and unlock modular reuse.

CTEG is the bridge between strategy and execution in intelligent systems.

About the Author

PAK is a hands-on engineering leader with a track record of building products and teams across Web3, AI, and gaming startups. Known for his obsessive attention to structure, he thrives in chaotic environments by applying systems thinking, frameworks, and operational playbooks to drive clarity and speed.

A firm believer in human-AI collaboration, PAK sees generative AI not as a replacement for talent, but as a force multiplier — freeing builders and thinkers to focus on what truly matters.

🔧 Explore the AI tool Pak is building — [join the waitlist here].

Composable Task Execution Graphs (CTEG): A New Foundation for Agentic Execution was originally published in Technology Playbook on Medium, where people are continuing the conversation by highlighting and responding to this story.

I. Introduction: Defining the Core System

At the heart of human inefficiency lies a systemic mismatch — not of access, but of alignment. We have more information than ever, yet the friction to apply it remains high. To understand why, we turn to systems thinking.

In this view:

• Knowledge Gap is a stock — the accumulated deficit between what someone knows and what they need to know to act effectively in context.
• Knowledge Transfer is a flow — the process by which information moves through networks, individuals, and systems.
• When knowledge transfer is effective, the knowledge gap decreases.
• If transfer is blocked, delayed, or poorly contextualized, the gap widens.

This forms a feedback loop:

• A smaller gap leads to better outcomes, which create new information and capacity for future action — fueling more knowledge transfer.
• A larger gap creates confusion, delay, and error — reducing engagement and feedback, reinforcing the deficit.

The Contextual Knowledge Gap Theory (CKGT) is the result of first-principles thinking about where the real constraints lie — not in generation, but in application. It repositions generative AI as a system-level tool for resolving the core inefficiency in human coordination: misaligned contextual knowledge.

II. Historical Milestones in Knowledge Flow

Human progress has followed breakthroughs in information transfer, as traced in Nexus, which documents how the structure and flow of information networks have shaped every stage of civilizational development:

• Language: Enabled intergenerational communication — laying the foundation for culture and progress through verbal knowledge transfer (Nexus, Ch. 1–2).
• Agriculture: Allowed dense populations to share skills and coordinate work through local and temporal knowledge networks (Nexus, Ch. 3), reinforcing the idea that information networks are prerequisites for societal complexity.
• Semiconductors: Powered computation — accelerating the speed and scale of knowledge processing across domains, creating self-reinforcing feedback loops between hardware and thought (Nexus, Ch. 7).
• The Internet: Liberated access to global knowledge, yet required users to manually extract, filter, and synthesize meaning. As Nexus notes, this network amplified reach but not necessarily relevance — and also introduced new risks around misinformation and asymmetric visibility.
• User-Generated Content: Created massive linguistic corpora — becoming the foundation for modern generative AI models. These decentralized knowledge inputs mirror the emergent, bottom-up dynamics emphasized throughout Nexus.
• Generative AI: Automates language processing — but still lacks alignment to individual goals, roles, and real-time context, and remains susceptible to algorithmic biases embedded in its training data and optimization goals.

Each step improved the flow of knowledge — but none solved the gap in meaningful contextualization. CKGT identifies this as the final constraint: the delivery of the right knowledge, to the right person, at the right time, in the right context.

III. Four Critical Use Cases for CKGT

1. Education

Decades of research suggest that if instruction could be fully personalized, learning efficiency could improve by over two standard deviations — a finding famously demonstrated in Bloom’s “2 Sigma Problem.” Yet scalable personalization has remained elusive. Learners today still struggle under cognitive load, where the inability to link new information with prior knowledge blocks progression. This is particularly acute in one-size-fits-all classrooms, where instructional pacing rarely matches individual understanding.

CKGT reframes this as a failure of contextual knowledge delivery. The learning gap isn’t just informational — it’s systemic. Cognitive Load Theory explains that excessive mental effort caused by poorly structured or misaligned instruction overloads working memory and prevents learning from taking hold. Generative AI has the potential to dynamically diagnose these gaps and deliver just-in-time, just-for-you content, precisely aligned with each learner’s cognitive state — finally addressing a pain point that has persisted for centuries.

2. Business

Enterprises consistently suffer from execution delays, misaligned handoffs, and siloed decision-making. These gaps in institutional knowledge lead to costly errors, stalled initiatives, and strategic drift. More critically, they threaten business continuity, especially during employee onboarding, off-boarding, or organizational change — when role-specific knowledge is fragmented or lost entirely.

CKGT interprets these challenges as manifestations of transfer loss — when what needs to be known to act is not where it needs to be, when it needs to be. Generative AI systems can encode, translate, and adapt knowledge across roles, functions, and time — turning business processes into living knowledge flows. By applying CKGT-aligned systems, businesses can preserve institutional memory, accelerate onboarding, and ensure consistent, context-aware execution across teams.

3. Commerce

Consumers today must navigate an abundance of product information, yet often struggle to make confident purchasing decisions. This is due to an inherent conflict: sellers describe their products in the best possible light, while user reviews reflect scattered, often subjective experiences. But even reviews can’t resolve the mismatch — because every consumer’s context and expectations differ.

A highly-rated coffee machine on Amazon may disappoint a café owner; a glowing TripAdvisor review may fail a traveler with different priorities. CKGT identifies this as a contextual trust gap. The failure lies not in the lack of data, but in the absence of personalized meaning. Generative AI can bridge this by aligning product information and reviews with the specific needs, preferences, and use cases of the buyer — restoring agency and trust in commerce.

4. Influencer Economy

The creator economy has become increasingly centralized and extractive. Platforms commoditize influencers by funneling them into ad-driven engagement models, reducing deep experiential knowledge into surface-level content optimized for clicks. The result is undervalued insight, distorted incentives, and growing platform dependence.

CKGT frames this as platform-induced knowledge gating — a systemic failure to deliver creator knowledge to the right audiences under fair terms. Algorithmic biases further amplify the problem, privileging content that performs, not content that informs. As discussed in Nexus, unchecked network effects and biased information flows can entrench asymmetries, deepen fragmentation, and distort public discourse.

Generative AI — when deployed as an agent of knowledge contextualization — can flip this paradigm. Such systems can empower creators to share contextual knowledge directly with aligned communities, enabling monetization through expertise rather than reach. This repositions influencers not as entertainers in an ad marketplace, but as peer educators in a democratized knowledge network.

Closing Thought

CKGT is not just a theory — it’s a systems model that reveals where generative AI’s real leverage lies: not in content creation, but in closing the contextual knowledge gap. In systems terms, knowledge gaps are stocks — accumulated deficiencies between what a person knows and what they need to know to act. Knowledge transfer is the flow that closes that gap. When flows are well-timed and personalized, feedback improves outcomes, and outcomes reinforce better flows — a self-reinforcing loop of learning, execution, and trust.

Generative AI, powerful as it is, is not sufficient on its own. It can produce endless outputs — but without context, that breadth creates noise, not clarity. Ask for a fried chicken recipe, and AI may generate 100 versions. But which one fits your dietary restrictions, your cultural palate, your available ingredients? Ask for a pizza recommendation, and you may get options ranging from Domino’s (speed) to artisan pizzerias (craft). But which one matters to you right now?

Even a product like Coca-Cola — with thousands of comparable substitutes — succeeds not because it’s the only option, but because the choice is pre-curated, contextually aligned, and consistently delivered. CKGT argues that this alignment is not incidental — it is the critical infrastructure for meaningful AI experiences.

Without understanding the user’s context, even the best model becomes a library without a librarian. With CKGT, we move from generic intelligence to contextual relevance — and that is where human impact begins.

In the coming weeks, I’ll publish a series unpacking how CKGT applies in detail across:

• Education
• Business
• Commerce
• The Influencer Economy

Each post will break down the system dynamics, the role of generative AI, and how CKGT can rewire the feedback loops that define each domain.

About the Author

PAK is a hands-on engineering leader with a track record of building products and teams across Web3, AI, and gaming startups. Known for his obsessive attention to structure, he thrives in chaotic environments by applying systems thinking, frameworks, and operational playbooks to drive clarity and speed.

A firm believer in human-AI collaboration, PAK sees generative AI not as a replacement for talent, but as a force multiplier — freeing builders and thinkers to focus on what truly matters.

🔧 Explore the AI tool Pak is building — [join the waitlist here].

The Contextual Knowledge Gap Theory (CKGT) was originally published in Technology Playbook on Medium, where people are continuing the conversation by highlighting and responding to this story.

TL;DR

• “The more decisions you have to make, the more exhausted you feel…. the less real work that gets done.”
• My MVT: Arc, Cursor, Notion
• Share your MVT here.

Tools, tools, tools

Every week I spent 10+ hours watching and learning new tools from YouTubers and online tutorials — mostly AI tools — they usually get me all excited about the new possibilities, hoping that one of the new tools will revolutionize my workflow.

While some tools are impressive, 9 out of 10 times the gain is marginal at best. One of the YouTuber, Bill Rice Strategy, said it well:

“Sam Altman, CEO of OpenAI, said those who move things forward aren’t the ones tweaking their system all day. They’re the ones who just choose the most important thing and do it.

Most productivity hacks are just disguised procrastination. We built fancy task managers, color code our calendars, and feel productive. But we’re not actually making progress. So here’s the truth. The more complicated your system, the more time you waste maintaining it. The more decisions you have to make, the more exhausted you feel, the more urgent task you chase, the less real work that gets done. And this is why the best productivity system is brutally simple.”

As a developer, I used to rely on Google, StackOverflow, and my fellow engineers to unblock me. Since 2024, Cursor and ChatGPT fill in 90% of all my knowledge gaps.

I removed most apps on my laptop, and I simplified my tools to the bare minimum. Here are my minimum viable tools (MVT):

Arc (browser)

You can set up different workspaces (the bottom icons) to group your links together. The favorites on top allow you access to common tools for all workspaces. This is the best place to store all your tools, project links, and bookmarks.

Cursor

Coding is never the same again with tools like Copilot, Cline, and Cursor. While YMMV, Cursor works best for me. I use it to scan for virus the first thing I do after downloaded a new repo; it helps me to learn a large code base, and it helps me tweak all the frontend TailwindCSS.

To get the most out of it, knowing which model to use and when helps.

Notion

Notion needs no introduction. It’s the ultimate note-taking and database tool. It has the best customizable user interface that you practically can replace most softwares with it. Only item on my wishlist is for Notion to provide more integration capabilities to empower their ecosystem.

Notion is a blank canvas, it requires know-how to get the most out of it. Start with the templates that usually comes with the best practices.Notion Calendar is amazing, and I have yet to try out the email and meeting notes features.

I use the Business Dynamics framework to manage my entire life using Notion. I track all the “stocks” (see image above) and “flows” using databases in Notion.

Other Tools

In addition, there are 5+ chat and meeting clients I have to install for comms as people are using different apps, it is beyond my control.

There are also usual office documents, and native AI apps fill up my dock. These three apps, however, produce most valuable work for me.

What are your MVTs?

Share your anonymous responses here.

About the Author

PAK is a hands-on engineering leader who has been building projects and teams for various startups, mostly focused in Web3, AI and gaming spaces.

He is an “organization freak” who has OCDs in keeping things organized in the most chaotic and fast-paced industries, using frameworks and playbooks.

He believes that AI will make engineers’ lives — and everyone’s lives better as many of the mundane tasks can be done by AI now, so humans can work on more interesting challenges.

Check out the AI tool Pak is building. Join the waitlist here.

MVT — Minimum Viable Tools was originally published in Technology Playbook on Medium, where people are continuing the conversation by highlighting and responding to this story.

From a trusted group on Telegram, I responded to this message above. I received a meeting invite immediately. When I open the browser link, it says “could not join the meeting from the browser” and asked me to download the app.

Just like a phishing website, it looks so legit, until you look at the details. I clicked on link and this was the file downloaded.

The person on Telegram asked me to download it, too. And when I pointed out this is a phishing attack, they removed the meeting link. I reported to the group owner immediately.

The Takeaways

• Don’t believe strangers from groups, even in a trusted group.
• Stay diligent out there! Do not try anything from untrusted source!

About the Author

PAK is a hands-on engineering leader who has been building projects and teams for various startups, mostly focused in Web3, AI and gaming spaces.

He is an “organization freak” who has OCDs in keeping things organized in the most chaotic and fast-paced industries, using frameworks and playbooks.

He believes that AI will make engineers’ lives — and everyone’s lives better as many of the mundane tasks can be done by AI now, so humans can work on more interesting challenges.

Check out the AI tool Pak is building. Join the waitlist here.

If you are interviewing for an engineering position, beware! Do not run any random repos without checking it!

TL;DR
There are scammers trying to steal your wallet private keys, and perhaps other hacks, too, by asking you to run arbitrary code. Do NOT run a random repo on your computer!

The Full Story

LinkedIn is a great platform if you are looking for jobs. 80% of my previous jobs are through random recruiters on LinkedIn. And for engineering jobs, technical interviews and coding tests are the norm.

A typical technical interview lasts between 30 to 60 minutes, perhaps the 10 to 15 minutes is just a short technical skill evaluation before the real interview, I thought to myself.

When I conduct technical interviews, I prefer to talk to the candidates, before any technical skill tests. Soft skills are as important as the technical skills; and context matters, too. So what happened next raised multiple red flags.

First, I check the repo and browse around the code base. In one of the API files, I notice an unusual horizontal scroll bar.

As I scroll, a block of obfuscated code reveals. This is the first red flag.

Then, I’m told that the CTO may or may not be present for the interview. This is the second red flag. How can you interview a candidate without showing up to the interview?

To confirm my suspicion, I asked AI to analyze the obfuscated code block. And here’s what it finds out:

Then I recall, one of the developers in my network shared a similar story. They were told to run a coding test repo and instructed to connect their crypto wallet during the interview. The process appears to be normal, until later the wallet was hacked and real crypto assets were lost.

The Ending

To prevent this account to scam others, I confronted the scammer directly.

And shortly after my reply, I witness the profile gets deleted.

The Takeaways

• Do NOT run arbitrary code for any reason. Even if it’s coming from friendlies. Always use AI to scan the code for security threats.
• Use a special purpose email and phone number for your resume/CV. You never know who is collecting your personal information.
• Always be alert for any red flags — anything out of the norm — when you are dealing with strangers. Safe job-hunting!

About the Author

PAK is a hands-on engineering leader who has been building projects and teams for various startups, mostly focused in Web3, AI and gaming spaces.

He is an “organization freak” who has OCDs in keeping things organized in the most chaotic and fast-paced industries, using frameworks and playbooks.

He believes that AI will make engineers’ lives — and everyone’s lives better as many of the mundane tasks can be done by AI now, so humans can work on more interesting challenges.

TL;DR

• AI implementation returns time — the most valuable resource
• Deploy using three core components: AI (“Artificial Intelligence”): role-based AI assistants, BI (“Business Intelligence”): proactively triggered actionable insights, CI (“Continuous Innovation”): iterations through feedback loops
• Implement with four key components: Vault (Private or Cloud), Queryable Knowledge Base, AI Tools, AI Assistants
• Strategic rollout using football analogy: Offense (growth), Defense (efficiency), Special team (ptimization and alignment)

Everyone recognizes AI’s transformative potential. But the practical question remains: How do you implement AI effectively with measurable ROI?

Give Back Time

Technology breakthroughs consistently deliver two critical benefits: quality improvements and time savings. And it is true with the AI revolution as well — it gives back the most precious resource — time!

The How-to Guide: AI, BI, and CI

AI

Deploying AI in your organization is straightforward — equip each team member with their own knowledge assistant!

Framing AI as a subordinate rather than a replacement reduces resistance some might perceive, while instantly boosting quality and productivity. AI can already help with conducting research, idea and content creation, image manipulation, coding, etc.

Even individual contributors can start delegating tasks to AI assistants and iterate with their professional feedback to reach the best solution. Each team member becomes both a curator and a decision-maker.

To maximize AI’s effectiveness, feed it your data. While AI contains vast general knowledge, it lacks your proprietary information. Provide organization-specific and role-specific knowledge to transform it from generic to specialized. Optimizations and improvements can be achieved through self-learning and capturing user feedback. With advanced memory features, AI assistants can adapt to your communication preferences over time.

BI

AI supercharges Business Intelligence (BI). Rather than relying on human experts to interpret dashboard visualizations, AI can analyze visualizations or data independently, deliver actionable insights, trigger timely alerts, and identify anomalies before they become problems, 24/7.

CI

“Continuous Innovation” requires structured feedback loops that distinguish effective from ineffective approaches. Both improvements and innovations emerge through measurable experiments with clear metrics.

Playbook

Standard Operating Procedures (SOPs) document essential processes for every aspect of your business. Collectively, these SOPs constitute your business playbook.

If you can document it, you can automate it. From lead generation to video editing, AI creates sophisticated automations previously impossible. Traditional automation required predefined outcomes with corresponding actions. With advanced reasoning, AI makes intelligent decisions independently, enabling truly autonomous operations.

Implementation

Effective AI deployment requires four distinct components:

1. Vault: Central repository for company knowledge with role-based access control. Typically combines cloud storage and databases.
2. Knowledge Base: Queryable intelligence with appropriate access restrictions. Usually built on a graph database with query engine, powering role-specific AI assistants.
3. AI Tools: Large Language Models and specialized tools customized for each organizational role.
4. AI Assistants: Conversational interfaces that understand specific roles, equipped with relevant tools and SOPs.

Rollout Strategy

Where should you rollout AI in your organization? Using a football analogy, focus on three areas:

Offense

Deploy AI to drive revenue growth through improved reach and conversion, personalization at scale, and enhanced customer experiences.

Defense

Boost operational efficiency and workforce productivity across all business functions.

Special Teams

Create internal alignment to optimize costs, products, and services for maximum impact.

Conclusion

Today’s businesses face unprecedented competitive pressure. Can you afford to ignore AI’s transformative potential? Or will you leverage AI to gain strategic advantage over your competitors?

About the Author

PAK is a hands-on engineering leader who has been building projects and teams for various startups, mostly focused in Web3, AI and gaming spaces.

He is an “organization freak” who has OCDs in keeping things organized in the most chaotic and fast-paced industries, using frameworks and playbooks.

He believes that AI will make engineers’ lives — and everyone’s lives better as many of the mundane tasks can be done by AI now, so humans can work on more interesting challenges.

Check out the AI tool Pak is building. Join the waitlist here.

Level up your business with AI — the only AI playbook you need was originally published in Technology Playbook on Medium, where people are continuing the conversation by highlighting and responding to this story.

Our data-analytics partners — NFTGo and Mystic — will select winners by analyzing Ethereum and Solana, respectively, and we will get the Twitter data from Hive.one.

LuckyBuy is the first on-chain SBT-based lucky draw Web3 infrastructure and application provider that empowers projects to reward their communities fairly and transparently.

As our event sponsor, LuckyBuy will be the exclusive prize-drawing partner for both on-chain and off-chain prizes. The whole process will be conducted on-chain, with prizes valued at over $100,000!

Our sponsors include ChachaSwap, Chamcha, Chief Toad, Mental College, EtherPOAP, TreasureLand, NFTPlay, and more to come!

There will be around 20 awards presented during this invite-only, Academy-style event. Notable guests from the NFT and Web3 space are invited to present the awards.

Event Highlights

• Over $100,000 worth of prizes
• 40+ blue-chip NFT communities
• High-value NFTs and sort-after Whitelists giveaway
• Hermes mystery boxes
• Free luxury hotel stays and nightclub access in Las Vegas

Investors, influencers, and NFT communities are invited to join the event for media coverage, interviews, networking, and, most importantly, having fun!

Sponsors & Partners

Inc DAO accelerates ideas in Web3 and links offline communities for T3 NFT Awards 2022.

0x499 is the most active decentralized community that provides exclusive online community partnerships for T3 NFT Awards 2022.

ChaChaSwap is the first community-driven, multi-chain, and royalty-supported decentralized NFT AMM protocol with a mission to improve the liquidity efficiency of the NFT market.

Chamcha makes buying, selling, and storing NFT Assets a simple and secure process.

Chief Toad is a collection of 10,000 unique NFTs. Your Chief Toad NFT doubles as G-link’s membership which gives VIP Access to G-Link’S gameverse platform.

Mental College- NUOS (Not Ur Ordinary School) is the first release in a multi-collection NFT series that takes place in a world where teenage students with mental disorders are chosen for their special capabilities and immersed into an expansive story arc. Mental College-NUOS is aiming to shape itself as the preeminent fashion/lifestyle platform in Web 3.0.

EtherPOAP is a series of community events led by the Ethereum ecosystem’s most ardent supporters and community members. Incubated by IPrinter DAO, EtherPOAP is dedicated to commemorating the historic Ethereum Merge and its subsequent upgrades.

TreasureLand, A multichain platform for NFT issuance, trading, NFT collections, and tailored in-shop services.

NFTPlay is a brand new NFT project that uses the interaction between devices and software technology to enrich the offline display and social ecology of NFT and enhance the utility of NFTs.

OKX is an innovative cryptocurrency exchange with advanced financial services. We rely on blockchain technology to provide everything you need for wise trading and investment. Enjoy hundreds of tokens and trading pairs. With OKX, you can join one of the leading crypto exchanges by trading volume. We’re serving millions of users in over 100 countries. We’re providing spot, margin, futures, options, perpetual swaps trading, DeFi, lending, and mining services.

Event Program *

• 6:30 pm — 7:30 pm Arrival/Networking
• 7:30 pm — 8:00 pm Lucky Draw #1
• 8:00 pm — 10:00 pm Awards Ceremony and Lucky Draw #2
• 10:00 pm — Midnight DJ/After Party and Lucky Draw #3

* subject to change

Please contact hello [at] t3event.com if you are interested in becoming a sponsor, presenter, media partner, or attendee.

Exclusive Data-Driven NFT Awards Ceremony was originally published in T3Event on Medium, where people are continuing the conversation by highlighting and responding to this story.

Hail, Guildmates!

My name is Calvin, and I’m part of BitGuild’s engineering team. There’s a lot more to building a full blockchain gaming ecosystem than you’d think, so I’m here to give you a snapshot of the steps involved in creating complex features such as our new Marketplace.

BitGuild Marketplace

As a blockchain gaming portal, BitGuild provides players with commonly used features, such as user management, language preferences, social sharing, game promotions, analytics, pre-sale, and marketplace — all so developers can focus on refining their core gameplay, and integrate with BitGuild to instantly get all the other features.

Some of our partners have already developed in-game marketplaces, but not all have implemented the ERC721 token standard that is used for creating game items. While it is easier to develop an in-game marketplace without adhering to the standard, the benefits of complying with them are definitely worth the extra effort. The most obvious benefit is being able to reach more traders on third party marketplaces.

In the first part of our marketplace deep dive, I’ll cover marketplace design decisions and the ERC721 callback — an important feature that is often overlooked. We will cover how to implement buy-with-PLAT and deal with encoding and decoding bytes in Part (II)

“While it is easier to develop an in-game marketplace without adhering to the standard, the benefits of complying with them are definitely worth the extra effort. The most obvious benefit is being able to reach more traders on third party marketplaces.”

Open Source

The BitGuild marketplace beta is open source, you can read the entire Solidity source code on GitHub.

Since it’s a beta release, we expect bugs to exist and we want your help to find them! Join the bug bounty program and help us find what we’ve missed.

Marketplace Design Decisions

1. Popular marketplaces only accept ETH as payment at the moment. On BitGuild, we need to also support PLAT — our official ERC20 token.
2. We considered storing the entire transaction history in the smart contract in an array or mapping. However, since Solidity does not handle large arrays well—and the more complex the code, the higher the gas price is—we’ve decided to go lean. The BitGuild marketplace smart contract only stores current listings. Transaction history is recorded via transfer events on the blockchain anyways.
3. Another debate we had is whether or not to allow users continue to use the game token when it is listed for sale. While it may sound like a good idea, it has unforeseeable consequences. What happens if a user lists the item on several marketplaces? And what happens if the item is sold while the seller is using it in-game? Because of this, the BitGuild marketplace requires game tokens be transferred to the marketplace smart contract when listed.
4. To uniquely identify each item listed for sale, there are two parameters: game contract address and the token ID. There are two approaches to this:
   - Chained mapping:
   mapping(address => mapping(uint => Listing) public listings;
   - Hashed key mapping:
   i) hash the game contract address and token ID:
   byte32 key = keccak256(abi.encodedPacked(_contract, _tokenId);
   ii) and use the key in the mapping:
   mapping(byte32 => Listing) public listings;
   Both approaches are essentially the same: “hash and then hash again.” But what if we have 3 parameters? The chained mapping will turn ugly quickly — imagine if we need two addresses and a uint, the mapping will look like: mapping(address => mapping(address => mapping(uint => Listing) public listings; Hence, we decided to go with the hashed key mapping.

ERC721 Callback

The ERC721 Callback hasn’t been discussed much. And if you Google it, the details could be well hidden in some lengthy articles about the full implementation of the ERC721. But without the callback, life would be much harder!

Let’s take a real life example: How do we list an item for sale?

Life without ERC721 callback
Intuitively, we will assume listing is simple — call a listing method on the marketplace contract, passing the game contract address, token ID and the price: MarketplaceInstance.list(contractAddress, tokenId, price); Since the token ownership info is stored inside the game contract, the marketplace contract can call the game contract transfer method to authorize the transfer of the token from the user to the marketplace contract: ERC721(contractAddress).safeTransferFrom(msg.sender, address(this), tokenId);

However, when the game contract receives the call above, the caller, or the msg.sender to the game contract is the marketplace contract, not the user. The token is still owned by the user. So this transfer is not authorized. It will fail.

The only way we can transfer the token to the marketplace contract, is if the user calls the game contract: GameContractInstance.safeTransferFrom(user.walletAddress, marketplaceAddress, tokenId);

Then we need to make another call to the marketplace smart contract to provide the listing details: MarketplaceInstance.listItem(contractAddress, tokenId, price);

So to list the item, we will need two transactions, and two gas fees. But the blockchain is distributed — we do not want to send the listing details without first having the user transferred the token to the marketplace contract. To make the calls happen sequentially, we need to manipulate the nonce (we’re won’t be covering this here).

But what if the token transfer fails? We cannot submit the second transaction until the first transfer is successful, so we’ll have to check the transaction status and initiate the listing using raw transaction only when it succeeds. That is a lot of code for implementing a listing function!

Listing with ERC721 Callback
What is the ERC721 callback? As the ERC721 standard suggests:

A wallet/broker/auction application MUST implement the wallet interface if it will accept safe transfers.

interface ERC721TokenReceiver {
    function onERC721Received(address _operator, address _from, uint256 _tokenId, bytes _data) external returns(bytes4);
}

And the callback is implemented inside the safeTransferFrom call:

function safeTransferFrom(address _from, address _to, uint _tokenId, bytes data) {
   // ...... handles transfer
   bytes4 retval = ERC721TokenReceiver(_to).onERC721Received(_operator, _from, _tokenId, data);
   require(retval == 0x150b7a02);
}

In the marketplace contract, add a handler for the callback:

function onERC721Received(address _operator, address _from, uint _tokenId, bytes _extraData) external returns(bytes4) {
    // ...... decode _extraData, handle listing
    return 0x150b7a02;
}

With the ERC721 callback, now listing an item can be done in one transaction:

GameContractInstance.safeTransferFrom(
  user.WalletAddress,
  marketplaceAddress,
  tokenId,
  encode(price) // encode extra data to bytes
);

Here’s the entire flow: a seller calls the safeTransferFrom method of the game contract to transfer the token to the marketplace contract, and the safeTransferFrom method will call the marketplace contract with the ERC721 callback so the listing info can be stored on the marketplace contract. All in just one transaction.

Notes: Older version of the ERC721TokenReceiver has no _operator field. To support older games you may want to support both interfaces.

What’s Next?

We hope this post was useful for tou! In the second part of this post, we’ll cover buy-with-PLAT implementation and how to deal with encoding and decoding bytes. Stay tuned!

BitGuild’s mission is to revolutionize the global gaming industry by creating a platform for a brand new class of games that live on the blockchain. Blockchain games completely redefine the relationship between players and developers by facilitating full and true ownership of in-game assets, cheap & safe item trading, cross-game compatibility of items & currency, and more.

Join the community on Twitter, Discord, and Facebook.

A Deep Dive Into Marketplace Smart Contracts was originally published in The Notice Board on Medium, where people are continuing the conversation by highlighting and responding to this story.