Productivity on TECHFOR by Suriya Sonphu

Develop a Software Development team: Beyond Speed to Strategic Velocity

Fri, 16 Jan 2026 00:00:00 +0000

In an era of rapid technological flux, where AI increasingly permeates daily life and significantly impacts the business sector, organizational leaders frequently confront a complex set of challenges: “How do we drive the organization forward while sustaining profitability and growth?”, “How do we keep pace with competitors?”, “How do we maintain our competitive edge?”, “How do we foster employee engagement, ownership, and skills, while simultaneously accelerating workflow?”, and crucially, “What is our AI strategy?” to expedite the software development process. From the adage of yesteryear, “big fish eat small fish,” to the subsequent era of “fast fish eat slow fish,” today, speed alone suffices not. One must possess the direction to guide the organization forward with both speed and orientation—Velocity. For if an organization moves too slowly, or rapidly but in an erroneous direction, it invariably leads to business failure.

1. Speed vs. Velocity

Itamar Friedman, CEO of Qodo, offered a simple yet potent definition: “The difference between Velocity and Speed is that Velocity has a vector.”

Focusing solely on speed makes the team appear perpetually busy, releasing new features frequently (High throughput), but these activities may not generate positive impact. Research titled “Speed vs Direction: Why Moving Fast Isn’t Always Progress,” Innovation and Strategy, Ibid. indicates that nearly 70% of Digital Transformation projects fail because execution precedes strategy. For large enterprises or organizations in highly regulated industries, rushing to code as fast as possible without quality assurance can lead to costly errors (AI-driven software development: Navigating the shift from speed to velocity, LinearB Blog).
Velocity is progress with purpose. For large organizations, Context Management and precision are essential. Thus, evolving from “vibe coding” to clear structural planning is paramount.

2. Defining via Vision and Strategy

To achieve Velocity, the team must first determine “Where do we want to go?” before asking, “We need to adopt AI; how do we execute this quickly?”

Vision is the long-term future image, akin to a North Star guiding the way. Strategy is the plan detailing how teams, systems, and processes will collaborate to realize that vision. Leaders must communicate this vision clearly to inspire and provide definite direction to the team (10 Strategic Thinking Techniques Used by Global Leaders, KCT Academy Thailand).

A tool aiding direction definition is the North Star Metric Framework (NSM), a key indicator reflecting the core value customers derive from the product. Having an NSM ensures all organizational parties (Product, Engineering, Marketing) work in alignment, focusing on the same outcome rather than working in silos (The Ultimate Guide to the North Star Product Framework, GeeksforGeeks).

3. Focus on Outcome over Output

Building software is not about increasing throughput; it is about injecting value into that throughput (Deliver Value).

In a case study on teams focusing on impact from the article Stop Obsessing Over Development Velocity, Focus on This Instead, Itamar Gilad notes that in experiments, teams that reduced throughput but spent time on Product Discovery and focused on the Success ratio generated better business results than teams focusing on throughput by fourfold.

Jeff Patton, a product expert, stated, “Your job isn’t to build more software. It’s to build less software (Minimize output) so that you can maximize outcome and impact.”

4. Techniques for Fast and Accurate Decision Making

Having direction does not imply slowing down. Research titled Making Fast Strategic Decisions in High-Velocity Environments by Kathleen M. Eisenhardt states that executive teams in the microcomputer industry found that teams deciding quickly and efficiently exhibit these behaviors:

Use Real-time data: They track operational metrics like daily bookings and cash flow closely, rather than relying solely on future forecasts.
Consider Simultaneous Alternatives: This helps visualize strengths and weaknesses comparatively and immediately reduces risk.
Employ a conflict management process: Fast teams use “Consensus with Qualification”—attempting to find consensus, but if unattainable, the leader makes a decisive call based on team input to avoid aimless waiting.

5. Balancing Tactical and Strategic Mindsets

Developers and team leaders must adapt their mindset appropriately.

Tactical: Focusing on the immediate task and solving immediate problems to complete the work.
Strategic: Focusing on the big picture, the future, and problem prevention.

Excessive tactical work leads to a “Crisis management loop” and burnout, while excessive strategic work can cause “Analysis Paralysis.” Good leaders must maintain this balance (The Strategic Vs. Tactical Mindset, DEV Community).

It is evident that software team development in the digital age is not a coding speed race, but a competition of who can “learn and adjust direction” towards creating customer value faster. Speed is merely an accelerator, but Velocity (Speed+Direction) is the true strategic asset. Successful organizations are those that cease running after currents, choosing instead to define clear goals and sprint towards them with agility.

Chain of Thought (CoT) prompting

Fri, 26 Dec 2025 00:00:00 +0000

Chain of Thought (CoT) is a technique in Prompt Engineering designed to enhance the reasoning capabilities of Large Language Models (LLMs), particularly for complex tasks. Instead of asking the model to provide an answer immediately, this technique instructs the model to display its thinking process step-by-step before reaching a final conclusion. This approach helps reduce errors caused by the model attempting to answer instantly without planning the problem-solving process.

Types of CoT Prompting

Key usage patterns identified in resources include:

Zero-shot CoT: Instructing the model to “think step-by-step” (e.g., using the command “explain your answer step-by-step”) without the user needing to provide examples of thinking beforehand.
Few-shot (Exemplar-based): Providing examples (Exemplars) that demonstrate how to break down a problem into steps, allowing the model to mimic that reasoning style.
Automatic CoT: The system automatically generates reasoning steps to reduce the burden on the user in writing prompts.
Multimodal CoT: Expanding the scope of reasoning by using multiple forms of data together, such as text and images, to solve more complex problems.

Benefits and Advantages

Higher Accuracy: Significantly improves the model’s ability to solve mathematical problems (Arithmetic reasoning) and logical reasoning tasks.
Transparency: Showing the thinking steps helps users understand how the model arrived at an answer, making it easier to debug errors.
Enhancing Smaller Models: Research suggests that CoT can enable smaller or cheaper models to have reasoning capabilities comparable to larger, more complex models.

Limitations and Considerations

Resource Costs: Generating multi-step thought sequences requires more processing power and takes longer, impacting costs.
Prompt Quality: The effectiveness of CoT depends on the quality of the prompts and examples provided. Poor prompts can lead to plausible but incorrect reasoning (Misleading reasoning).
Risk of Overfitting: The model might memorize the response patterns from examples too closely, leading to a lack of flexibility in solving diverse problems.

Real-world Applications

CoT is widely used in various fields, such as:

Education: Creating detailed explanations for solving math or science problems.
Customer Service: Helping chatbots understand complex issues and provide more targeted answers.
Research: Assisting researchers in structuring thought sequences for scientific problem-solving.
AI Ethics: Explaining the reasoning behind AI decisions in situations where ethics must be considered.

Comparison: To visualize this clearly, Chain of Thought is like a student showing their work on a math exam instead of just writing the final answer. Showing the work ensures that each calculation step is correct, and if the final answer is wrong, the teacher (or user) can trace back to see where the error began.

Example: Applying CoT in Product Development Strategy

1. Step-by-Step Prompting

Instead of asking “How should we develop a hotel booking app to beat competitors?”, use Instruction-based CoT by ordering the model to separate its thinking into a sequence:

Step 1: Analyze the Pain Points of users in the current market.
Step 2: Compare the Pros and Cons of key competitors in the market.
Step 3: Identify key features (MVP) that can solve problems better than competitors.
Step 4: Plan a development Roadmap prioritized by impact on users.

2. Using the Thinking Process for Transparency

Using CoT helps the development team understand the “origin” of a strategy. For example, if the model suggests emphasizing an AI Chatbot feature, CoT will make the model show supporting reasons (Reasoning steps) such as:

“Analysis indicates that users often cancel bookings if they don’t receive an answer within 5 minutes.”
“Therefore, adding a Chatbot will help reduce the Churn rate and save on customer service costs.”
Result: This strategy is logically more sound than just suggesting a feature name.

3. Applying Various CoT Patterns (Variants)

Zero-shot CoT: Users can use short commands like “Strategize expanding the app’s user base by thinking through it step-by-step in detail” to let the model use its internal knowledge to build a thought sequence.
Few-shot (Exemplar-based): Users provide examples of successful past strategies along with the thinking behind them, so the model mimics that reasoning pattern for a new product.
Multimodal CoT: If there are app UI mockups, images can be used with text for the model to analyze and propose UX strategy improvements based on visual data.

4. Benefits for Digital Product Development

Reduced Errors: Prevents jumping to strategic conclusions that might not fit the actual market (Avoiding jumping to an answer).
Deep Problem Solving (Multistep Reasoning): Helps visualize the connection between technology (Backend) and business needs (Business Goal) more clearly.
Precision: Breaking a large problem into small parts (Manageable steps) helps make the plan detailed and actionable.

Using CoT in digital strategy is like a team whiteboarding session where we don’t just talk about the final goal, but we sketch out the sequence of events, from understanding the customer to selecting the technology, so everyone sees clearly why we chose that strategy.

Chain of Thought (CoT) prompting is not just a technique for writing commands to get answers, but a crucial mechanism for elevating the reasoning potential of AI to be closer to human thought processes.

Transforming AI from an “Answerer” to a “Thinking Partner”: The core of CoT is shifting from letting the model jump to what seems like the correct answer (jumping to an answer) to displaying a transparent, sequential problem-solving process, which significantly reduces reasoning errors.
Creating Transparency & Observability: Having the model “think out loud” in steps allows users to verify errors and understand the logic behind the AI, making it less of a Black Box.
Expanding Capabilities Without Limiting Model Size: CoT allows smaller models to process complex tasks like math problems or ethical decisions similarly to larger models.
The Trade-off: Despite the benefits, users must trade off higher computational costs and longer processing times, as well as the risk of generating plausible yet incorrect reasoning if prompts are not rigorous enough.

Therefore, using CoT in work, whether for business strategy or technical problem-solving, is like “forcing AI to show its work on scratch paper.” This not only helps obtain more accurate answers but also helps us as users learn systematic problem analysis methods along with the AI.

References

Microsoft Learn. (2025, May 29). Chain of Thought Prompting - .NET. Retrieved from https://learn.microsoft.com
- Key Focus: Focuses on using Prompt Engineering to display sequential steps and intermediate results to reduce model errors.
Coursera Staff. (2025, April 18). Chain of Thought Prompting: Enhancing AI Reasoning and Decision-Making. Retrieved from https://www.coursera.org
- Key Focus: Explains types of CoT (Zero-shot, Automatic, Multimodal) and its application in arithmetic and commonsense reasoning tasks.
Gadesha, V., Kavlakoglu, E., & Winland, V. (IBM Research). What is chain of thought (CoT) prompting? Retrieved from https://www.ibm.com
- Key Focus: Details advantages regarding transparency and limitations concerning high computational power costs.

Our Learning Path is Not Linear: Understanding the Learning Journey

Sun, 07 Dec 2025 00:00:00 +0000

Have you ever asked yourself why, when we see successful people, we often only see them standing at the peak, but never the scars, scrapes, or worn-out shoes they endured on their journey up?

We are often instilled with a “linear path” mindset: if we want to excel at something, whether it’s learning a new programming language, managing a team, or even practicing a sport, the process should be simple: Start -> Learn -> Practice -> Succeed.

But when we actually start doing it… why isn’t it like that? Why are there moments of discouragement? Why are there times when no matter what we do, it doesn’t get better? Or are we just not good enough?

The short answer is… “No, you’re not inadequate. It’s just that the map you’re holding is wrong.”

Today, I’m here to lay out the “Reality” map for you.

The Learning Path: Myth vs. Reality

If you look at the infographic above, you’ll see a clear difference.

Left side (The Myth): This is what we expect. We want the graph of life to shoot up in a straight 45-degree line, beautiful and predictable.
Right side (The Reality): This is what actually happens. It’s messy, winding, and goes in circles, more like a maze than an escalator.

I want to invite everyone to understand the 8 crucial stages you “must encounter” in real life, so we can prepare and deal with them like leaders.

Delving into the 8 Stages of Learning (You Can’t Avoid)

1. The Spark

Everything begins with passion or some necessity. During this period, our energy is at its highest, like when you’ve just started a new project or bought a new book.

2. The Struggle

Once you actually start… “Wait, this isn’t what we talked about!” Difficulties begin to appear, self-doubt sets in. This is the first critical point where most people “give up.”

3. The Learning Loop (Trial -> Learn -> Retry Loop)

This is the core. In the real world of work (especially in tech), we need to Fail Fast.

Do: Start coding, start managing.
Learn: See what the results are (they are often imperfect).
Retry: Use what you’ve learned to improve.

This loop is not a failure; it’s about “tuning” the frequency to match the goal.

4. Failure as Feedback

Change your mindset: scars are not a mark of inadequacy, but data points telling you, “This way isn’t working. Try another path.”

5. The Plateau

This is the most important and often the most boring stage. It’s when you’re doing everything right, working just as hard, but the results seem stagnant. Many people quit here, thinking they’re on the wrong path.

But the truth is, “you are accumulating latent potential.” It’s like water boiling; it looks still until it suddenly bubbles vigorously at the last minute.

6. The Pivot

When old methods no longer move us forward, we must be brave enough to stop, reassess, and change strategy. Stubbornness isn’t always perseverance; rather, flexibility is the key to survival.

7. Small Wins

Don’t wait to celebrate only at the finish line. Cherish the pride along the way. Successfully coding one function, completing a difficult task—these are excellent fuels.

8. Success as a New Beginning (The Goal as a New Start)

When you reach the peak, you’ll discover the truth: there’s a higher peak waiting ahead. Learning is not a “destination”; it is a “lifestyle.”

Embrace the Mess

As someone who works in development and management, I can confidently say that no project is as smooth as a bed of roses, and no skill is mastered overnight.

At the finish line, everyone might look the same—“successful.” But what distinguishes a “professional” from an “amateur” is the ability to endure through stage 5 (The Plateau) and the courage to get back up during stage 3 (The Loop).

So, if your learning journey today is messy, confusing, or feels like you’re going in circles… Congratulations, you’re on the right track!

As long as you don’t stop walking, this winding path will lead you to the most beautiful places.

And you? Where do you think you are stuck on this map right now? ^_^

WindRecorder | Personal Memory Search Engine

Fri, 11 Jul 2025 00:00:00 +0000

What is WindRecorder?

WindRecorder is an open-source Windows application that functions as a Personal Memory Search Engine - an alternative to tools like Rewind (for Mac) or Microsoft Copilot Recall. It records everything happening on your screen in small file sizes, allowing you to rewind what you’ve seen, search through OCR text or image descriptions, and get activity statistics.

Key Features of WindRecorder

1. Intelligent Screen Recording

Records multiple or single screens, or just the active window
Smaller file sizes with lower system resource usage
Stable, continuous capture with real-time rewind capabilities

2. Smart Data Indexing

Only indexes changed scenes and updates OCR text, page titles, browser URLs to the database
Custom skip conditions by window title, process name, included text, or screen idle time
Automatic database maintenance, video compression and cleanup when computer is idle

3. Complete Web Interface

Review screens, conduct OCR and image semantic searches
Multi-language support: Simplified Chinese, English, and Japanese built-in

4. Data Analytics & Insights

Activity statistics, word clouds, timelines, light boxes, scatter plots
AI-powered tag summarization using LLM

5. Multiple OCR Engine Support Beyond Windows’ built-in OCR capabilities, it supports various third-party OCR engines:

Rapid OCR: onnxruntime version of Paddle OCR
WeChat OCR: Extremely high accuracy for Chinese and English recognition
Tesseract OCR: Supports 100+ languages with simultaneous multi-language recognition

What Are the Benefits of WindRecorder?

1. Privacy & Security

100% Offline Operation: No internet connection required, no data uploads
You Own Your Data: All data stored locally on your computer only
Complete Control: You decide what to record and when

2. Performance Efficiency

Small File Sizes: 2-100 MB per hour (depends on screen changes/number of monitors)
Low Resource Usage: Minimal impact on system performance
Automatic Maintenance: Cleans and compresses data during idle time

3. Search Convenience

Text Search: OCR-powered search through any text that appeared on screen
Image Search: AI-powered content analysis of visual elements
Timeline Navigation: Easy browsing through historical activities

Using WindRecorder for Programming Work

1. Work Tracking & Documentation

Scenario: You're debugging a complex issue and need to retrace your steps
Solution: Search for keywords like "error", "debug", "console" in WindRecorder
Result: Find screenshots showing error messages and debugging steps you took

Scenario: Need to create documentation or tutorials from past work
Usage: Search by project name or technology stack used
Benefit: Get actual screenshots and workflows for documentation creation

3. Troubleshooting & Debugging

Scenario: Previously working code now has issues - need to see what changed
Workflow:
1. Search by filename or function name having issues
2. Review timeline of changes during specified periods
3. Compare before/after states of the problematic code

4. Project Management & Time Tracking

Scenario: Need to report work progress or analyze time spent on tasks
Usage:
1. View time statistics across different applications
2. Analyze work patterns from word clouds and timelines
3. Use data to optimize work efficiency

5. Knowledge Backup & Research

Scenario: Research gathering from multiple sources while problem-solving
Benefits:
- Record documentation pages you've visited
- Capture code examples from various sources
- Build personal knowledge base from daily work

Real-World Example: API Development

When developing an e-commerce API:

Error Log Tracking: Search for “500 error” or “database connection failed”
Database Schema Changes: Search by table names that were modified
API Documentation Review: Find API docs you previously accessed
Performance Analysis: Review time spent on API testing activities

Summary

WindRecorder is a powerful tool for developers and general computer users who want to create their own digital memory system.

Advantages:

Privacy-First: All data stays on your machine
Efficient: Low resource usage with high value output
Flexible: Multi-language support with various search methods
Open Source: Free and customizable to your needs

Considerations:

Windows Only: Currently no Mac or Linux versions available
Early Development: May encounter minor issues during use
Storage Requirements: Approximately 10-20 GB per month (depending on usage)

For developers who need tools to help manage knowledge and work experience, WindRecorder is an interesting alternative, especially for those who prioritize data privacy and security.

Additional Resources:

This article introduces WindRecorder, an open-source tool that helps developers and general users manage their digital memory efficiently and securely.