MIT study (Jul 2025): only 5% of AI pilots in companies actually create real financial value.

McKinsey study (Aug 2026): only 6% of companies are what they call "AI high performers."

Two different teams. Two different ways of researching it. Different sample size. But almost the same number.

This is not a coincidence. This is telling us something real.

And here is the part that is worrisome. McKinsey in 2026 report found 37% of companies say AI touched their earnings "a little". That number is basically flat compared to one year before (McKinsey published same report in 2025).

Meanwhile companies invested a lot more money on AI in that same year and rolled it out to many more teams. So more money invested, more adoption, but the return line is staying almost the same.

80% of employees still say AI makes them personally more productive. Only 37% of companies can point to real impact on earnings ("a little"). Everyone feels faster. Companies feel happier about imaginary ROI. But the P&L is not catching up, and it has not caught up in a full year, staying around mid-single digit (5-6%).

What is separating the small group who succeed from everyone else?

• Buy the tool, don't try to build everything yourself. Companies working with outside partners succeed two times more than the ones building in-house.

• Change how you work, not just add AI on top of the old process. The companies doing well are three times more likely to redesign their workflow completely.

• Put money in the places like operations and back-office, not only sales and marketing demos. This is where the real ROI is hiding.

• Leaders need to be personally involved. Not just leave it to the IT team, AI engineers or the data scientists.

• The tool needs memory. If it forgets everything each time, people will stop using it, quietly.

So as always, technology (AI in this case) is not really the problem here. Maybe it is us, and how we are using/adopting/leading/implementing it.

References MIT study: https://webberwentzel.com/News/Documents/2025/MIT1757411281972.pdf

McKinsey study: https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai#/

A recent study by Carnegie Mellon and Stanford, analyzing 196,212 pull requests (PRs) at one company, found that per-developer PR creation increased to 2.09× the baseline, while the review workload per reviewer nearly doubled.

In other words, AI has made developers significantly more productive, but it has also created a growing queue of PRs that still depend on human reviewers. Several industry reports are pointing to the same conclusion.

Are AI Code Review Agents the missing piece?

We may now be seeing the next wave of AI engineering tools: AI Code Review Agents.

One of the most interesting examples comes from PostHog (Product Analytics and Engineering Company), which recently published an engineering article titled "Stop Being the Code Review Bottleneck."

Their internal AI code review agent, StampHog, has already given the final approval for roughly one-third of the PRs independently handling approximately 1,600 PRs.

For perhaps the first time, we're seeing an engineering organization allowing an AI agent to approve a meaningful share PRs.

What makes their approach interesting?

Rather than replacing human reviewers, the goal is to automate routine reviews and allow engineers to focus on other high-value engineering work.

Some notable engineering practices adopted by PostHog include:

• AI reviews routine PRs and escalates only higher-risk changes.

• The agent that generates the code never reviews its own work.

• Large changes are broken into smaller, incremental PRs, making automated validation more reliable.

• Requirements and implementation intent are validated before code generation begins, reducing downstream review effort.

The real insight isn't that AI will replace code reviewers. It's that AI review and human review are complementary - not competitive.

This isn't an isolated experiment. GitHub recently announced that Copilot Code Review surpassed 60 million code reviews by March 2026 - a 10× increase since April 2025.

Of course, AI code review isn't free. Running AI code review agents introduces additional inference and token costs, so organizations will need to balance those costs against the engineering productivity they unlock.

I am curious to hear how other engineering teams are approaching this.

References

Posthog Article: https://newsletter.posthog.com/p/code-review-tips

Carnegie Mellon Stanford Study: https://arxiv.org/pdf/2607.01904v1

GitHub Code Review Growth: https://github.blog/ai-and-ml/github-copilot/60-million-copilot-code-reviews-and-counting

When Anthropic launched its autonomous, multi-agent Claude Code platform in February 2026, it felt like a massive breakthrough by solving deep repository ingestion and complex business context understanding and industry welcomed it with open arms.

But the honeymoon phase was short-lived. A new wave of operational bottlenecks quickly emerged. While a few organizations like Rakuten have managed to successfully scale their AI code generation, others are struggling to stay afloat against five new challenges:

Tokenmaxxing: Blowing through AI token budgets under the false assumption that more consumption equals more progress.

Review Bottlenecks: Human engineers spending hours in massive waves of AI-generated code.

Code Bloat: Unnecessary complexity and redundant lines injected into software architectures.

Security Exposure: Brand-new vulnerabilities introduced by autonomous agents operating without guardrails.

Junior Pipeline Collapse: A shrinking runway for junior developers to learn and grow, as AI swallows all the basic tasks.

Why are some organizations succeeding while others fail?

Let’s look into the most discussed issue in tech circles today: Tokenmaxxing.

📉 The Rise of Tokenmaxxing Tokenmaxxing stems from a fundamental misunderstanding - tech leaders assuming that consuming more AI data automatically equates to higher engineering productivity.

Companies like Uber and Microsoft learned this the hard way. Within months of deploying Claude Code, organizations realized that skyrocketing token usage didn’t guarantee higher-quality output. The financial reality hit quickly—AI token costs are no longer a temporary experiment; they are now a permanent, heavy operational expense.

🛠️ How the Industry is Fighting Back Engineering

Leaders must actively manage token budgets the exact same way they manage cloud infrastructure spend. To prevent cost overruns, companies are shifting from unrestricted AI access to structured token governance.

💡 Real-World Impact: Tech firm Branch8 recently published an audit detailing how they slashed their Claude Code daily API spend by 69%—all without sacrificing engineering velocity.

While every team's infrastructure looks different, successful organizations generally attack the problem across three main pillars. Here are a few actionable examples of strategies being used on the ground today:

Governance Frameworks (Policy & Enforcement)

• Budget Caps: Implement strict monthly AI spend limits per employee.

• Agent Restrictions: Fully disable expensive, built-in "Explore" or "Plan" subagents unless explicitly approved.

• Mandated Guidelines: Publish clear, company-wide AI usage compliance rules.

Cost-Control Tools (Automated Technical Levers)

• AI Gateways: Deploy centralized proxy gateways to monitor, throttle, and log AI requests.

• Prompt Caching: Utilize caching mechanisms to avoid paying for the same repository context over and over.

Developer Best Practices (Day-to-Day Habits) Lean Context

• Keep files like CLAUDE.md lightweight and highly relevant.

• Context Auditing: Train developers to inspect their context windows before running large queries.

• MCP Server Management: Regularly audit Model Context Protocol (MCP) servers to ensure they aren't feeding bloated data to the AI.

By treating AI tokens as a finite infrastructure resource rather than a blank check, engineering teams can finally stop burning budgets and start shipping better code.

🔗 References Rakuten Case Study: https://claude.com/customers/rakuten

Branch8 Token Optimization Audit: https://branch8.com/posts/claude-code-token-limits-cost-optimization-apac-teams

Uber AI Budget Trajectory: https://aimagazine.com/news/why-uber-has-already-burned-through-its-ai-budget

Here is how that looks on the ground:

1. The Demo vs. Reality Gap: It’s easy to build a cool AI demo in a weekend, but making it ready for the real world is a massive engineering challenge. The biggest risk is using AI to write code too quickly - it can create significant technical debt. While the code might work today, it can be difficult to fix or modify later.

2. The Legacy System Problem: AI models are powerful, but most companies are still running decades-old legacy systems with outdated databases and disconnected software. Execution involves the unglamorous work of re-architecting and redesigning these systems so that AI actually has something useful to work with.

3. The J-Curve of Productivity: Recent 2025-2026 data (referenced below) shows that when companies first introduce AI into a team, productivity often drops initially. Employees spend more time supervising the AI, verifying its work, and figuring out new workflows. Real execution requires the patience to redesign the job itself - not just sprinkle AI on top of existing processes.

Prominent examples of success and failure in AI initiatives

Success - Klarna Customer Service Transformation - Klarna, the Swedish fintech giant, is often cited as a gold standard for AI execution. They developed an AI assistant and completely redesigned their customer service workflow using OpenAI’s enterprise tools. The AI was deeply integrated into their payment and refund systems across 35 languages. In its first month, the AI handled two-thirds of all customer service chats - equivalent to the work of 700 full-time agents - while maintaining customer satisfaction scores comparable to human agents.

Failure - Air Canada Liability Risk - This is a classic example of poor execution of legal and policy guardrails. Air Canada deployed a chatbot to help travelers with FAQs. When a customer asked about bereavement fares, the bot hallucinated a policy, telling him he could claim a refund after his flight - something that contradicted the company’s actual policy. Air Canada argued in court that the chatbot was a separate legal entity and that the airline wasn’t responsible for its responses. The court rejected this argument, forcing the airline to pay the refund and damages.

The real challenge with AI is not potential - it is disciplined execution.

References

J-Curve Productivity: https://mitsloan.mit.edu/ideas-made-to-matter/productivity-paradox-ai-adoption-manufacturing-firms

Klarna Customer Service Transformation: https://www.klarna.com/international/press/klarna-ai-assistant-handles-two-thirds-of-customer-service-chats-in-its-first-month/

Air Canada Liability Risk: https://www.pinsentmasons.com/out-law/news/air-canada-chatbot-case-highlights-ai-liability-risks

When ATMs were introduced in the 1960s, many braced for the death of the bank teller. However, the availability of 24×7 access to cash led to increased consumer spending and demand for banking products. The bank teller’s job transformed to providing a variety of services. Since then for 40 years before the advent of digital banking, the number of bank teller jobs has grown faster than the overall labour force. Time and again, technologies initially feared as job killers have ultimately expanded opportunities.

Similarly, most modern-day jobs were unimaginable a century ago. When a good becomes more abundant, we find new ways to use it, and entirely new categories of work are created. History is filled with such examples - from coal consumption in the 19th century to data consumption in the 21st century. As William Stanley Jevons, the 19th-century English economist, observed: gains in efficiency often lead to increased consumption. This phenomenon is popularly known as Jevons Paradox.

Why Do We Fear It?

At the advent of any major technology, it is unclear how it will ultimately benefit individuals and corporations. The unknown naturally creates worry. However, that worry is often rooted in uncertainty rather than evidence. As technology matures and its applications become clearer, fear gradually gives way to understanding.

How Do We Overcome the Fear?

As in the case of bank tellers, who learned to provide a variety of new services as their roles evolved beyond dispensing cash, it is important to learn and adapt to changing realities. Embracing AI as an opportunity - for personal growth and organizational advancement.

SWE-bench is a widely followed benchmark evaluation framework designed to test AI coding assistants on real software engineering tasks.

AI coding assistant benchmarks are supposed to give us clarity. SWE-bench does the opposite.

SWE-bench-Verified has been widely considered a Python-language-only, 500-task benchmark, where the best models recently reached around an 80% score (roughly 400 out of 500 tasks solved). The industry celebrated

Then Scale AI introduced SWE-bench-Pro: harder tasks, multiple programming languages, and more realistic, real-world software engineering scenarios. Suddenly, the same frontier models crashed to around a 23% score 😬

But wait - it gets better.

Someone noticed that the 23% score was calculated using a “capped turn limit and capped cost”. Give the models more attempts and an uncapped budget? The best-performing models jumped to ~46%. Same benchmark. Same models.

Benchmarks keep shifting, evaluation rules keep changing, and scores keep flip-flopping. The industry needs standardized, stable evaluation protocols.

Before we benchmark models, we need to benchmark the benchmarks.

Reference: https://scale.com/leaderboard/swe_bench_pro_public

The term "Creative Destruction" describes the process through which new innovations, technologies, or business models replace older ones. In doing so, they dismantle outdated systems but simultaneously create new opportunities, industries, and wealth.

Key insights from the Nobel-winning work:

• Without the right culture, institutions, education, and incentives, creative destruction can stall or even backfire.

• New developed model helps economists analyze the conditions under which innovation thrives or falters and what policies can help sustain long-term growth.

These insights will guide governments and organizations in refining policies that encourage innovation while managing its disruptive effects.

Interestingly, nature itself operates on the same principle. From cellular regeneration to the cycle of seasons, destruction in nature is never waste it’s renewal. The ecosystem flourishes because it never clings to the obsolete. This perpetual cycle of death → transformation → rebirth keeps life diverse, adaptive, and resilient.

Over the past fifty years, we’ve witnessed a similar cycles in technology. Here too, creative destruction drives progress. Old systems, tools, and business models fade away as smarter, faster, and more efficient technologies emerge. This ongoing renewal fuels productivity, reshapes industries, and transforms how we work, communicate, and create value in the digital age.

Lesson: Just as nature has survived and thrived for millions of years through creative destruction, technology too will continue to reinvent itself, creating endless new opportunities for growth and innovation.

What the data actually says

- Devil in the detail: General-purpose LLMs are widely piloted and deployed (40%). Task-specific GenAI stalls only ~5% reach production.
- Good signal for Tech & Media: Despite the hype, AI shows up clear structural disruption mainly in Tech and Media so far out of 9 researched industries.
- Start simple: AI has already “won” simple work. 70% prefer AI for drafting emails, 65% for basic analysis. For complex, multi-week work, humans still dominate 9:1.
- Approach matters: The divide isn’t driven by model quality or regulation; it’s driven by workflow fit, learning, and memory.
- Where ROI hides: Early, measurable savings come from reducing BPO and external agency spend especially in back-office operations more than headcount cuts.

Playbooks that work for organizations (from the 5% who succeed)

- Land small, visible wins in narrow workflows; then expand.
- Choose tools with low configuration and immediate visible value. Initially avoid heavy enterprise customization.
- Prioritize learning, memory, and workflow adaptation not flashy UX.
- Partner to build pilots: AI pilots via strategic partnerships are 2x as likely to reach full deployment as internal builds.

Research Report Link: https://lnkd.in/g3UCZ8Nr

Problem Addressed: Manually fixing sanitizer bugs like memory corruption and thread issues is time-intensive and prone to human error. Automating bug fixes for software vulnerabilities detected by sanitizers in languages like C/C++, Java, and Go. Google developed an AI-powered patching pipeline using Large Language Models (LLMs) for automating bug fixes.

Google's AI-powered patching pipeline leverages a structured five-step process to automate bug fixing effectively.

1. Detect vulnerabilities: The pipeline identifies sanitizer bugs (errors like memory corruption or thread issues) and reliably reproduces them to ensure they are actionable.

2. Isolate bugs: It narrows the focus to the problemetic code section, enabling precise prompts for Large Language Models (LLMs).

3. Generate fixes with AI: Using an LLM, such as Google's Gemini, it crafts accurate code patches tailored to the problem.

4. Test proposed fixes: The pipeline automates the creation of commits from the generated patches, integrating them into the codebase and running extensive automated tests.

5. Human review: Even after passing all tests, the machine-generated patches undergo rigorous review by developers to ensure safety and functionality.

Observations:

• The system scales across large codebases, improving efficiency in handling bugs.

• This model can help fixing all kind of bugs not just sanitizer bugs.

• All patches undergo rigorous testing to ensure reliability before deployment.

• Automated the fixing of 15% of sanitizer bugs, translating to hundreds of successful patches.

• Fixes generated by AI are subject to human review, enhancing accuracy.

• Faster patching reduces security risks, minimizing exposure to exploits.

Reference: https://storage.googleapis.com/gweb-research2023-media/pubtools/7563.pdf

Secret Cyborg: Refers to individuals who rely on AI tools (like ChatGPT, Copilot, or DALL·E) to assist with their tasks without openly acknowledging it. They seamlessly integrate AI into their workflow, making them "cyborg-like" but secretly.

Digital Blacksmith: Creators or developers who use GenAI tools to forge new digital products, like virtual avatars, AI-assisted games, or synthetic media.

Prompt Chaining: The technique of breaking down a task into a series of prompts fed sequentially into an AI model to improve the quality or specificity of results.

Prompt Leaking: When confidential or sensitive information is accidentally revealed in an AI's training or response due to poorly controlled inputs.

Federated Learning: A decentralized approach to training AI models using data distributed across multiple devices while maintaining user privacy.

Synthetic Reality: Entirely AI-generated virtual environments or scenarios indistinguishable from real life.

AI Alignment Tax: The extra computational or developmental cost incurred to make AI systems safer and more aligned with human values.

Digital Twin: A virtual replica of a physical entity, created and managed by AI to simulate, predict, and optimize performance.

Zero-shot Learning: When an AI system performs a task without having been explicitly trained on it, relying instead on general knowledge.

Diffusion Model: A type of AI model used to generate images or other content by iteratively refining random noise until it forms coherent outputs (e.g., DALL·E or Stable Diffusion).

Foundation Model: A large-scale AI model, like GPT or BERT, trained on vast amounts of data and designed to serve as a base for a wide range of downstream tasks.

Model Alignment: The process of ensuring that an AI system's goals, outputs, and behavior align with human values, ethics, and intentions.

Explainability (XAI): A branch of AI focused on making AI models' decisions and processes understandable to humans.

Model Compression: Techniques used to reduce the size of an AI model while retaining performance, enabling its deployment on resource-constrained devices.

Synthetic Persona: AI-generated characters or personalities designed for specific purposes, such as customer service bots or virtual influencers.

AI Dungeon Master (AI DM): An AI system used in role-playing games to create dynamic storylines and control the narrative for players, often leveraging GenAI to generate plot twists and dialogues.

Neural Dust: Miniaturized neural interfaces powered by AI, often used to monitor or interact with biological systems like the brain or other organs.

Synthetic Media: Media (images, videos, audio, or text) generated entirely by AI, such as AI-created music, movies, or artworks.

AI Agents: Autonomous AI programs designed to perceive their environment, make decisions, and take actions to achieve specific goals.

AI Companion: An AI-based personal assistant or chatbot designed for companionship, advice, or emotional support, like Replika or Character.AI.

Model Fine-tuning: Adjusting a pre-trained AI model by training it further on a smaller, specialized dataset to optimize its performance for specific tasks.

AI Ethics: The field dedicated to studying and ensuring ethical considerations in AI development and deployment, addressing issues like bias, privacy, and accountability.

Prompt Engineering: The practice of crafting precise and effective prompts to guide AI systems for generating accurate, creative, or desired outputs.

Hallucination: When an AI system generates incorrect or nonsensical information confidently, despite no factual basis for its response.

Interview clip

The End of SaaS as We Know It

Here's a simplified explanation of Satya Nadella's idea, current state and future state diagrams for comparison.

Current State (SaaS-Based Applications):

• SaaS applications like CRM (Customer Relationship Management), ERP (Enterprise Resource Planning), and tools like Excel rely on CRUD operations (Create, Read, Update, Delete) for managing data.

• These applications have their own backend database, business logic, and frontend interface.

• Users interact with each application separately, and the logic (rules and workflows) is embedded in each application.

Future State (AI-Agent-Based Applications):

• AI Agents will act as intelligent middle layers. Instead of having business logic inside each application, the AI layer (agents like Copilot) will handle workflows, decisions, and data operations.

• These agents will not be tied to a single app or database but will interact with multiple systems seamlessly (e.g., combining data from multiple databases).

• This removes the need for separate SaaS applications as the AI Agent will centralize and orchestrate all business operations.

How AI Agents Will Eliminate SaaS Layers?

AI Agents are trained on vast amounts of data related to the business processes they are designed to manage. This data can include historical records from existing systems, industry best practices, and insights from human experts. By analyzing this information, AI agents identify patterns, relationships, and business rules, effectively building a comprehensive knowledge base. This knowledge base allows them to make informed decisions and take actions autonomously.

Once trained, AI agents leverage natural language processing (NLP) to understand user requests and respond in a conversational, human-friendly manner. These interactions can take place through chat interfaces, voice assistants, or even via automatically generated reports and summaries, offering users seamless and intuitive experiences.

Impact of the AI Agent Model

SaaS Providers:
SaaS providers will need to transition from building standalone applications to creating AI agents that integrate seamlessly with various systems and data sources. This shift requires expertise in AI/ML, data integration, and API development. Additionally, it presents opportunities for companies to develop AI agent platforms and tools, fostering an innovative ecosystem around AI agent technologies.

App Economy:
The rapid increase of AI agents may lead to app consolidation as these agents take over functions currently handled by individual applications. This evolution could introduce new business models, such as subscription-based services for AI agent access or pay-per-use pricing for specific AI-driven tasks, fundamentally reshaping the app economy.

Job Market:
While AI agents may reduce the demand for certain roles, such as front-end and back-end development for SaaS apps or chatbot creation, they will simultaneously create opportunities in emerging fields. These include AI development, Python programming, machine learning, deep learning, natural language processing (NLP), and data analysis. New professions focused on training, managing, and overseeing AI systems will become essential in this evolving landscape.

Conclusion

The rise of AI Agents signifies a transformative shift in how SaaS providers manage operations and interact with technology. By centralizing workflows and decisions, these agents have the potential to replace traditional SaaS applications, enabling more efficient, intelligent, and seamless experiences. This evolution challenges SaaS providers to adapt, reshapes the app economy, and redefines job roles, creating both disruption and new opportunities. As AI agents continue to mature, they promise to not only revolutionize the SaaS ecosystem but also redefine the future of technology-driven innovation.

• Mark Zuckerberg believes there could eventually be more AI agents than people in the world.

• Nvidia CEO Jensen Huang envisions AI agents playing critical roles across various sectors, including marketing and chip design.

What is Agentic AI?

Agentic AI is driven AI Agents operating on top of large language models (LLMs) and is an autonomous system that understands, thinks, and acts by processing given instructions. It transforms AI intelligence into actionable steps, following a sequence of instructions to perform tasks, and continuously learns through a cycle of instructions, processing, action, and improvement. In simple words its the next wave of AI built on top of Gen AI.

For example, an AI email assistant processes incoming emails to understand their context, drafts replies based on the content, and can even send responses autonomously.

Why Agentic AI is Needed?

Generative AI, in its current form, is bi-directional: users provide prompts, and LLMs generate responses. However, it cannot autonomously perform tasks, make decisions, or operate without continuous human intervention.

AI agents, on the other hand, work on top of LLMs and are designed to automate tasks, make decisions, and solve problems without constant supervision. They save time, reduce errors, and boost efficiency by handling repetitive or complex tasks, offering personalized assistance, and enabling smarter decision-making. When tens, hundreds, or even thousands of AI agents collaborate, they unlock massive scalability and productivity, transforming workflows across industries.

What is AI Agent's Potential in Software Engineering?

AI Agents have immense potential in software engineering, particularly within the Software Development Life Cycle (SDLC). Advanced AI Agents can work in sequence, where the output of one becomes the input for the next. This creates a multiplier effect on scalability, productivity, and efficiency.

For instance, starting with project requirements, AI agents can collaborate seamlessly to produce the final output, significantly streamlining the entire SDLC process. Feedback loops enable these agents to learn and improve over time, facilitating continuous improvement.

Example sequential workflow of AI Agents in SDLC: AI Agents can integrate advanced workflows to take business requirements as inputs and, after passing through various stages, deploy code to production while generating user documentation.

1. Requirement Analysis AI Agent

   • Extracts and refines project requirements from inputs like client documents, emails, or interviews.

   • Output: Structured and detailed requirements document or data.

2. UI/UX Design AI Agent

   • Creates prototypes, wireframes, and visual designs based on the requirements.

   • Output: UI/UX designs and design specifications.

3. Code Generation AI Agent

   • Converts UI/UX designs and requirements into working application code.

   • Output: Codebase with functionality matching the design and requirements.

4. Automated Testing AI Agent

   • Analyses the code, generates test cases, executes tests, and evaluates results.

   • Output: Test results, reports, and identified issues.

5. Bug Detection and Fixing AI Agent

   • Reviews test results, identifies bugs, and generates patches or fixes for the issues.

   • Output: Debugged and functional code.

6. DevOps Automation AI Agent

   • Automates CI/CD pipelines for building, testing, and deploying the application to staging or production.

   • Output: Deployed application or service.

7. Code Refactoring AI Agent

   • Improves code structure, readability, and efficiency without changing functionality.

   • Output: Clean, optimized code.

8. Documentation AI Agent

   • Creates technical documentation, API guides, or user manuals based on the final codebase and workflows.

   • Output: Complete project documentation.

9. Knowledge Sharing AI Agent

   • Summarizes the project, creates onboarding guides, and generates training materials for new team members.

   • Output: Knowledge base, training documents, or video tutorials.

Conclusion

Agentic AI is a big step forward in how we use artificial intelligence. Unlike regular AI, they can work on their own, learn, and handle complex tasks without constant help. By saving time and improving efficiency, AI Agents are set to transform industries like software development, making work faster and smarter.

These are six areas in Software Engineering where Generative AI can reliably and quickly deliver higher productivity gains. By prioritizing these areas, organizations can speed up the process of productivity gain and gradually venture into more complex tasks. These areas offer significant productivity gains with minimal time spent reviewing and implementing the generated output. Let’s explore how AI is making a tangible difference in each of these areas.

1. Basic UI Design

Designing user interfaces often involves converting textual descriptions or wireframes into functional code. Generative AI excels at this by rapidly producing HTML/CSS templates or React components based on simple instructions. For example, developers can describe their needs in plain language, and the AI delivers a working prototype.

This capability accelerates early development stages, enabling faster iteration and allowing teams to focus on refining designs instead of building them from scratch.

2. Test Case Generation

Writing test cases is a critical yet labour-intensive part of software testing, consuming 20% to 40% of testers’ time depending on project complexity. Generative AI reduces this burden by efficiently generating test cases with good accuracy. Although AI cannot replace human testers entirely, it aids in creating diverse and comprehensive test scenarios. This allows QA teams to focus on exploratory testing and edge cases, while AI handles repetitive or routine tasks.

3. Creating Documentation

Documentation is often a bottleneck in software development, taking up 10% to 20% of project time depending on complexity. Generative AI simplifies this process by creation of different documents with ease and accuracy. These documents can be:

• User guides and technical manuals

• Commenting and annotating code

• API documentation

With AI tools, developers can produce clear, consistent documentation effortlessly, improving team communication and stakeholder understanding. This saves time that can be redirected to coding, testing, or innovation.

4. Codebase Maintenance, Enhancement, and Support

In the software industry, roughly 70% of projects are focused on maintenance, enhancement, and support. These areas often offer the most significant productivity gains for developers using Gen AI. Recent empirical study published by Ness-Zinnov has established that Engineers witnessed maximum impact when utilizing existing codebase functions, leading to reduced development cycle time. Generative AI assists by suggesting improvements, identifying inefficiencies, or generating new functions using existing patterns, making it easier to navigate complex or unfamiliar codebases.

5. Code Refactoring

Refactoring, the process of improving code structure without altering functionality, is tedious yet essential for maintainability. Generative AI simplifies this by suggesting optimized structures and removing redundancies. This is particularly valuable for legacy systems, where outdated or poorly documented code can hinder innovation. With AI, teams can ensure cleaner, more efficient codebases, making debugging and future enhancements easier.

6. Data Analysis

Data analysis, a cornerstone of decision-making, involves examining, organizing, and interpreting data to uncover insights. Generative AI enhances this process by accurately analysing data and identifying patterns. For software teams, this enables faster analysis of log files, user behaviour data, or system performance metrics, allowing for quicker and more accurate decision-making.

Conclusion

Generative AI is a transformative force in software engineering, especially for tasks like UI design, test case generation, documentation, code maintenance, refactoring, and data analysis. By delegating these routine responsibilities to AI, developers and testers can focus on innovation, problem-solving, and building high-quality software. However, Generative AI is not without limitations. Its performance declines with increasing task complexity, and human oversight is still essential to ensure quality and correctness. For instance, AI-generated code or documentation may contain errors or inconsistencies that require manual review.

As the technology evolves, its reliability for complex tasks will improve, unlocking even greater potential. For now, teams that embrace Generative AI can gain a competitive edge through faster delivery, higher quality, and enhanced productivity.

These discrepancies raise important questions about the validity of GitHub's high productivity improvement claims. However, it is crucial to delve into the subsequent research to understand the potential reasons for these variations.

Mert Demirer, an assistant professor of economics at MIT Sloan, aptly noted that measuring how AI impacts productivity in real-world workplace environments is a significant challenge. In this context, understanding the factors contributing to the wide range of productivity outcomes in different studies on Copilot becomes essential.

Given that Copilot has been the leading AI pair-programming tool over the past two years, I reviewed 17 empirical studies, papers and articles (references available at the bottom) focused on its productivity impact. From this research, I identified six possible factors and three additional reasons suggested by ChatGPT that can help technology leaders, planners, managers, and developers better align their expectations to achieve the desired productivity improvements.

Six factors contributing to productivity variation

Six factors outlined in the studies collectively explain much of the variation in developer productivity when using GitHub Copilot. Here's how each contributes to the observed differences:

1. Developer Experience: Productivity varied by developer experience, with less experienced developers getting more benefit from Copilot. [Ref #8].

2. Adoption and Usage Patterns: Short-tenured and junior developers were more likely to adopt Copilot and to continue using it for more than one month, and that these developers were more likely to accept the output code generated by Copilot. [Ref #8]

Takeaways from factor#1 and #2:

• Organizations and teams with a higher proportion of junior developers can expect significant productivity gains from adopting GitHub Copilot. These teams may also benefit from reduced project delivery risks typically associated with less experienced developers, as Copilot's assistance helps bridge knowledge gaps and enhance code quality.

• For teams with a greater number of senior developers, expectations should be adjusted to reflect incremental productivity improvements, as experienced developers may already possess the expertise that Copilot enhances.

• Additionally, the adoption and sustained usage patterns among junior developers suggest that organizations can confidently onboard and deploy them on critical projects while leveraging Copilot to maintain or even exceed delivery expectations and reduce project cost.

3. Language-Specific Performance: Notable differences among the programming languages in terms of correctness of suggestions (between 57%, for Java, and 27%, for JavaScript). [Ref #11]

Takeaways:

• Recognize that GitHub Copilot's performance varies by programming language. When planning its adoption, assess the predominant languages used in your organization and set realistic expectations for productivity gains and code quality improvements.

• Prioritize investment in languages where Copilot demonstrates higher suggestion accuracy (e.g., Java).

4. Existing Codebase Utilization: Engineers witnessed maximum impact when utilizing existing codebase functions, leading to reduced development cycle time. [Ref #9]

Takeaway:

• Organizations and teams working on maintenance and support work can expect increased productivity benefits compared to the organisations starting projects from scratch.

5. Sampling Effectiveness: A study by OpenAI found that Codex (Copilot underlying model responsible for generating code) has a 29% effectiveness when using the first sample, 47% effectiveness when using the best of ten samples, and 78% effectiveness when using the best of a hundred samples. In other words, the more code samples you generate, the more likely one of them will pass the testing suite. [Ref #10]

Takeaways:

• Tech leaders should invest in training programs to teach developers how to effectively evaluate and test generated samples, ensuring they choose the most suitable option for their requirements.

• Developers should generate multiple samples when solving complex problems to increase the likelihood of finding a high-quality solution.

6. Task Simplicity: Copilot is accurate for simpler tasks. Worth exploring task decomposition for better accuracy. [Ref #12]

Takeaway:

• Developers should use Copilot with confidence for simpler tasks, repetitive patterns, or standard logic implementations, saving time for more complex challenges and breaking down complex tasks into smaller, well-defined steps before attempting to generate code with Copilot. This not only increases accuracy but also helps you better understand the problem at hand.

Additional factors contributing to variations in productivity

Beyond the above outlined reasons revealed through various studies and research papers, other factors can also influence productivity. These additional reasons are generated by ChatGPT when inquired about the reasons for Copilot productivity variations.

1. Learning Curve: Developers new to Copilot may initially experience reduced productivity due to the time needed to learn how to use it effectively.

Takeaways:

• Organisations should plan workshops or training programs to familiarize developers with Copilot’s features, workflows, and best practices.

• Recognize that developers new to Copilot may experience a short-term productivity decline. Set realistic expectations for adoption timelines and provide support during the learning phase.

2. Code Quality Review: The need to verify and refactor AI-generated code can sometimes offset productivity gains, particularly in critical or high-stakes projects.

Takeaway:

• Organizations should emphasize that AI tools like GitHub Copilot are not replacements for human judgment. Integrate their use with established code review practice, coding standards and incorporating code review and quality tools such as SonarQube, DeepSource etc.

3. Ethical and Security Concerns: Concerns about insecure or unoriginal code might lead to increased scrutiny, reducing overall efficiency.

Takeaway:

• Organisations need to encourage developers to validate Copilot's output for originality and security, and establish clear accountability for code quality by setting up following security protocols, coding-standards, and using the tools such as Veracode, Checkmarx, Fortify, Black Duck etc.

These factors, collectively, offer a comprehensive explanation for the observed variations in productivity when using Copilot.

Conclusion

Understanding the factors behind GitHub Copilot's varied productivity outcomes is crucial for organizations and developers looking to maximize its benefits. While Copilot holds great potential, its effectiveness depends on several variables, including developer experience, task complexity, language-specific performance, and code review practices. By addressing these factors and aligning expectations, teams can better leverage Copilot to enhance efficiency. Additionally, navigating the learning curve, improving sampling strategies, and addressing concerns around security and code quality are essential for unlocking its full potential. This nuanced understanding can help technology leaders, planners and managers and developers make informed decisions, fostering greater innovation and productivity in software development.

References

#1. Research-quantifying-github-copilots-impact-on-developer-productivity-and-happiness

#2. Is-github-copilot-worth-it-real-world-data-reveals-the-answer

#3. Measuring-github-copilots-impact-on-engineering-productivity

#4. Ai-for-developer-productivity

#5. The-impact-of-github-copilot-on-developer-productivity-a-case-study

#6. The Productivity Effects of Generative AI

#7. Ebay-generative-ai-development

#8. copilot-developer-productivity

#9. generative-ai-improves-software-engineering-productivity-by-70-says-ness-zinnov-study

#10. Increasing-Productivity-With-GitHub-Copilot

#11. On the Robustness of Code Generation Techniques

#12. Evaluating the Usability of Code Generation Tools

#13. Is GitHub Copilot a Substitute for Human Pair-programming?

#14. how-generative-ai-affects-highly-skilled-workers

#15. An Empirical Evaluation of GitHub Copilot’s Code Suggestions

#16. Measuring-github-copilots-impact-on-productivity

#17. Productivity Assessment of Neural Code Completion

AI and the Fear of Job Loss

AI is poised to revolutionize industries, automating routine and repetitive tasks. This will inevitably lead to the displacement of certain roles. However, the fear of tech-induced unemployment is not new. When personal computers emerged in the 1970s, many feared they would render clerical workers obsolete. Similarly, the rise of the internet in the 1980s and mobile phones in the 1990s sparked fears of job losses in traditional industries.

Yet, these innovations fueled the global economy. For example, while automation reduced manufacturing jobs, it simultaneously created millions of roles in IT, software development, and digital marketing. In 1970, the world had approximately 1.6 billion jobs; by 2020, this figure had grown to 3.5 billion, driven by new industries and technologies. The digital revolution, powered by the internet, gave rise to professions like web developers, app designers, and cybersecurity experts, which were unimaginable a few decades earlier.

Here is the quick snapshop of growth in global jobs with rising population (Source: United Nations and World Bank).

Similarly, AI is expected to create new economy jobs in areas such as AI development, data analysis, robotics maintenance, and ethical governance. A World Economic Forum report predicts that while 85 million jobs may be displaced by AI by 2025, 97 million new roles will emerge, representing a net gain.

The Need to Unlearn and Relearn

Preparing for an AI-driven future requires a willingness to adapt. The most valuable skill in the age of AI is the ability to unlearn outdated methods, learn new technologies, and train in emerging fields. For instance, as AI automates routine accounting tasks, accountants who upskill in data analytics and financial forecasting will remain indispensable.

Governments, businesses, and educational institutions play a crucial role in this transition. Workforce development programs, reskilling initiatives, and affordable access to education can help individuals navigate the changing job landscape. At the same time, individuals must take ownership of their learning journeys, embracing lifelong learning and cultivating skills in creativity, critical thinking, and emotional intelligence—areas where humans excel over machines.

The Bigger Picture

Rather than fearing AI, it is crucial to view it as a tool for economic growth and productivity. By automating repetitive tasks, AI allows humans to focus on innovation and problem-solving. This shift can lead to higher wages, better working conditions, and improved quality of life.

History shows that while old economy jobs may fade, the overall number of jobs continues to grow. The key lies in adaptability. By embracing change and investing in skills for the future, individuals and societies can harness AI’s potential to create a more prosperous and equitable world.

In conclusion, AI will transform the world of work, just as personal computers, the internet, and mobile phones have done before. The future of jobs depends not on resisting change but on learning to thrive in it.

I was experimenting with the Square Web Payment SDK available on GitHub. After downloading the SDK, I completed the basic configuration in the .env file using NodeJS, including setting the ACCESS_TOKEN. When I executed the code for card_payment.html on port 3000, I encountered an authentication error. This issue puzzled me because running the code seemed straightforward, and I expected it to work without any issues.

Determined to resolve the problem, I, being a novice NodeJS developer, spent almost an hour troubleshooting the issue on my own. I double-checked the configuration settings, server.js file, ensured that the APPLICATION_ID and LOCATION_ID were correctly set to my project, and verified that the SQUARE_ACCESS_TOKEN was accurately updated in the .env file. Despite these efforts, I continued to receive the "Payment failed" error. The detailed error message in the console indicated an AUTHENTICATION_ERROR with the code UNAUTHORIZED, suggesting that the request could not be authorized.

Feeling perplexed, I turned to ChatGPT for assistance. I explained my issue to ChatGPT using the following prompt:

Prompt: I am using the SQUARE web-payments-quickstart SDK available on this GitHub URL "https://github.com/square/web-payments-quickstart.git". In the SDK, there is an example code for different modes of payments available in this folder "web-payments-quickstart/public/examples". I am using card-payment.html to make the card payment. I have changed the APPLICATION_ID and LOCATION_ID in my project. I have also created a .env file and updated "SQUARE_ACCESS_TOKEN" in the file. However, when I am running "card-payment.html" on port 3000, I am getting a "Payment failed" error when clicking "Pay $1.00". Please suggest what can be the issue. I guess it may be an issue with "SQUARE_ACCESS_TOKEN" not being used anywhere in the "server.js" file.

This is the detailed error message in the console: [ [Object: null prototype] { category: 'AUTHENTICATION_ERROR', code: 'UNAUTHORIZED', detail: 'This request could not be authorized.' } ]

This is the server.js file code [server.js code was provided here]

ChatGPT diagnosed the issue in the first instance and suggested that the problem might be related to the SQUARE_ACCESS_TOKEN not being used correctly in the server.js file. ChatGPT generated an updated server.js file. I didn't even review the code and replaced the existing server.js code with the newly generated code. With these changes, I was able to successfully test the Square Web Payment SDK, resolving the authentication error and completing the card payment process smoothly.

GitHub Link: https://github.com/rajpachori/SquarePaymentSDK

Empowering Novice Developers

One of the most striking benefits of GitHub Copilot is its ability to empower novice developers. As someone without any prior experience in React Native, I approached the task with a mixture of excitement and apprehension. By spending sometime on writing a detailed prompt for Copilot Chat, I was able to create a functional application that handled CRUD operations (Create, Read, Update, Delete) for employee records, all driven by a NodeJS server component returning JSON data.

Accelerating Development Time

Time is a precious commodity in the world of software development, and assistant programming tools like GitHub Copilot excel at improving productivity and saving cost. The entire process, of successfully writing the code, took a mere 30 minutes. Copilot's ability to understand the context of my code and provide exact suggestion on the first try played a crucial role in this rapid development cycle. This acceleration not only boosts productivity but also allows developers to focus on more complex and creative aspects of their projects.

The Prompt: Employee Management App

A detailed prompt is crucial for maximizing the benefits of assistant programming tools like GitHub Copilot. It provides clear context and specific requirements, enabling the AI to generate accurate and relevant code suggestions. This clarity accelerates the development process, ensures functionality aligns with user needs, and enhances the overall quality of the generated code, making it a vital component for efficient programming.

Detailed prompt used for generating the required code

Use index.tsx and create new program from scratch for searching, creating, updating and deleting the employee. NodeJS server component is successfully deployed and working fine on port number 4000. NodeJS returns the information in JSON format.

These will be the UI components.

1. Text "input" field for entering the EmpID.

2. "Search" button along side Text input field for searching the employee.

3. Table with 2 columns. First column will have below mentioned 7 field names and second column will have values for 7 fields displayed in Input text field. a. EmpID b. EmpName c. Designation d. EmpDOJ e. ManagerName f. Compensation g. EmpStatus

4. "Create", "Update" and "Delete" button below the table.

This is the expected behavior of each button

1. "Search" button: Data fetched from JSON is populated in the second column "input" field of the table beside respective field name.

2. "Create" button: Data available in each of the input fields, creates a new employee data.

3. "Update" button: Data available in each of the input fields, updates employee data for EmpID value populated in the second column of the table.

4. "Delete" button: Delete employee data for EmpID value populated in the second column of the table.

Conclusion

Assistant programming tools like Copilot are not just aids; they are game-changers that democratize software development, enabling anyone with a passion for coding to bring their ideas to life swiftly and effectively. As I continue my journey in the world of AI assisted development, I look forward to leveraging these tools to explore new horizons and understand how these tools can help in improving productivity, saving cost and maximize revenue.

The website offers distinctive desktop wallpapers and backgrounds for Windows users, with wallpaper images generated using DALL-E, an AI program. Thus, I embarked on an innovative project to create https://wally123.com, a streamlined, single-page website. Utilizing GitHub Copilot, I could successfully build and deploy the website in just 8 hours, demonstrating the remarkable efficiency of AI-assisted programming in accelerating web development tasks. It took me approximately 3 hours to program the website, with the remaining time spent on purchasing a domain and hosting plan, generating roughly 20 images using DALL-E, and deploying the website.

GitHub Copilot's role in this development process was pivotal. It offers more than just code completion; it acts as an intelligent assistant, suggesting whole lines or blocks of code based on the project's context. This enables it to anticipate coding needs, recommend best practices, and even suggest alternative problem-solving approaches. For instance, when I needed to implement a feature for dynamically displaying the AI-generated wallpapers, Copilot provided snippets with well-explained comments that seamlessly integrated JavaScript with the HTML structure, significantly reducing complexity and coding time. Additionally, the Copilot Chat feature can assist in fixing bugs, explaining code, testing code, and more.

GitHub Copilot’s ability to interpret comments and convert them into code proposals also enhanced the development workflow. By simply describing a functionality in plain English within a comment, Copilot would propose relevant code, enabling rapid iteration on new ideas and features without getting bogged down in syntax details.

- Click here to check the source code of the website

In conclusion, Copilot can significantly improve developer productivity and efficiency, aid in debugging, explain complex code, generate unit test cases, and perform many more complex tasks as an AI pair-programming tool.

- Click here to read research conducted on productivity improvement with the help of GitHub Copilot

The Initial Challenge

Having a good experience in object-oriented programming (OOP) and experience in C#, I was familiar with the fundamental concepts of programming. However, I didn't have any hands-on experience on Android development and Java. The idea of creating an app from scratch seemed daunting, especially given my lack of experience in this specific area. That’s when I turned to ChatGPT for assistance for implementing an idea of developing “Greetings” App.

Learning Prompt Engineering

As I was new to ChatGPT, I first wanted to learn the basics of prompt engineering so I could efficiently use ChatGPT for my experiment on programming. I took a few basic prompt-engineering courses on YouTube and spent around three hours on the same.

Setting Up Android Studio and Basics

The first step was setting up my development environment. I needed to install Android Studio and configure it. I had known a bit about Android Studio earlier, so it took around one hour, and I didn’t really need any help from ChatGPT for setting it up. I also had some understanding of the core components of an Android app, such as Controls, Activities, Layouts, and Resources from my prior experience managing the development teams.

Implementing Functionality (This is where ChatGPT was most helpful)

I had a clear understanding of the expected functionality from “Greetings” App, so I started by defining the core features my app needed. I first wanted to ensure that clear functional requirements were documented in a step-by-step process and used as a prompt in ChatGPT, which took me around one hour. This, I believe, is the most important step in getting the intended and quality output from ChatGPT.

In my interactions with ChatGPT, I learned how to handle user inputs, code events & functions, and navigate between different screens. For instance, ChatGPT guided me through setting up event listeners for buttons, enabling them to respond to user actions. It provided clear code snippets of how to write the corresponding Java code, explaining the logic step-by-step.

However, when I applied the code in Android Studio and built it, the code threw build errors multiple times. It was not a cakewalk to simply copy-paste the code for the application to start working. I had to interact with ChatGPT multiple times before getting the code ready and making it bug-free. However, each time I received a build error, I turned back to ChatGPT for solving the error and ChatGPT broke down the concepts into manageable parts. Each time I hit a roadblock, ChatGPT offered immediate solutions and alternative approaches, allowing me to maintain steady progress. At one point, ChatGPT was not able to resolve a problem related to a screen layout, and after trying different approaches, it could not produce a workable solution so I had to change the screen layout a bit in expectation of resolving the issue, which ChatGPT successfully did on the first try.

Overall, writing and debugging the core functionality took about ten hours. The real-time feedback and tailored explanations from ChatGPT were crucial in transforming my high-level understanding of programming into practical Android development skills. By the end of this phase, I had a functional app that responded to user interactions as intended.

Below some screenshots from the app

Source code link on GitHub

Points to Note

• I used ChatGPT-4 and wrote 90% of the code with its help.

• Before using ChatGPT for coding, it’s important to clearly understand the requirement and frame the prompt.

• The quality of responses improves drastically by providing detailed and clearer prompts.

• ChatGPT is not perfect. It can produce faulty code which can be debugged by further conversations.

• I used Midjourney and Canva for generating the images required for my app. The time taken in learning Midjourney and generating the images is not included in the 15 hours.

Conclusion

Building my first Android app in just 15 hours (including learning basics of prompt engineering) was a remarkable experience. ChatGPT proved to be an exceptional guide, bridging the gap between my existing programming knowledge and the new skills required for Android development. For anyone hesitant to embark on a new technical venture, I recommend leveraging the capabilities of ChatGPT — it might just surprise you how quickly you can achieve your goals.

Image Editing

1. PhotoRoom: AI-powered tool for removing backgrounds and editing photos easily.

2. PIXLR: Online photo editing tools powered by AI for quick and professional edits.

3. cutout.pro: AI tools for background removal and photo enhancement.

4. Hotpot: AI-powered design tools for image editing and graphic creation.

5. Leonardo: AI-driven platform for generating, editing, and enhancing digital art and content.

Video/Audio Editing

6. VEED.IO: AI-driven video editing platform for creating, editing, and sharing videos.

7. Clipchamp: Online video editing platform with AI features for easy video creation.

8. runway: AI tools for creative professionals, including video editing and image synthesis.

9. KAPWING: Online video editing platform with AI-powered tools for easy creation.

10. VocalRemover: AI tool for removing vocals from songs to create instrumental tracks.

Text to Speech

11. Speechify: AI-powered text-to-speech tool for converting text into spoken words.

12. ElevenLabs: AI-driven text-to-speech and voice cloning services for realistic voices.

Writing and Content Generation

13. Poe: AI-powered platform for generating written content and creative ideas.

14. QuillBot: AI writing assistant for paraphrasing, summarizing, and improving writing.

15. TheB.AI: AI-powered writing and content generation tool for various purposes.

16. Writesonic: AI-powered writing assistant for generating high-quality content.

17. copy.ai: AI-powered writing tools for generating marketing copy and content.

18. ZeroGPT: AI-powered content generation platform for unique and high-quality text.

19. Smodin: AI tools for writing, translation, and content generation in multiple languages.

20. WRITER: AI-powered writing and editing tools for high-quality content creation.

21. NovelAI: AI-assisted storytelling tool for generating and enhancing narratives.

22. GPTGO.ai: AI tools for generating creative content and writing assistance.

AI Art and Image Generation

23. CIVITAI: Community-driven platform for generating and sharing AI-created images.

24. nightcafe: AI art generator for creating unique and visually appealing artworks.

25. Lexica: Search engine for discovering and exploring AI-generated images.

26. Midjourney: AI-generated art platform creating unique, high-quality visual artworks.

27. PIXAI: Platform for creating and discovering AI-generated art.

28. Stability: AI model for generating realistic images from text descriptions and enhancing digital creativity.

29. Looka: AI-powered logo maker and branding platform for custom logo designs.

AI-Powered Search Engines

30. YOU: AI-powered search engine offering personalized search results and summaries.

31. Perplexity: AI-powered search engine providing detailed answers and explanations from multiple sources.

Presentations and Storytelling

32. tome: AI-powered tool for creating visual storytelling and presentations.

33. Gamma: AI-powered presentation tool for creating visually appealing presentations.

Digital Humans and Deepfakes

34. D-ID: AI solutions for creating digital humans and privacy protection in media.

35. DeepSwap: AI-powered deepfake technology for creating realistic face swaps.

AI Development and Experimentation

36. replicate: Platform for running and sharing machine learning models in the cloud.

37. Playground: Platform for experimenting with AI models and building AI applications.

38. Hugging Face: AI community and platform for sharing, developing, and deploying machine learning models.

Marketing and Sales Automation

39. ZMO: AI solutions for automating marketing and sales processes.

Data Analytics and Business Intelligence

40. Palantir: Offering AI-driven data analytics solutions for government and commercial sectors.

Robotic Process Automation

41. UiPath: Leading in robotic process automation (RPA) with AI-driven automation tools.

Enterprise AI Solutions

42. C3.ai: Providing enterprise AI software for various industries, focusing on digital transformation.

43.ForefrontAI: AI solutions for automating customer interactions and business processes.

44. SenseTime: Specializing in AI and deep learning technologies, particularly in computer vision and facial recognition.

Others

45. character.ai: AI-generated character interactions and conversations for entertainment and utility.

46. ChatPDF: AI tool for interacting with and extracting information from PDF documents.

47. Chub.ai: Platform for creating and deploying AI-powered chatbots.

48. Fliki: AI platform for converting text into videos and voiceovers.