#19 AI Coding Assistants Deliver High Returns

Plus: AI Adoption among Asset Managers, Use Cases in Production, AI Tools to Try and Notable AI News

👋 The newsletter is back! I took a month off to travel in Japan and China. Thank you for your patience; I hope you had a fantastic summer with family and friends. This newsletter returns to its usual bi-weekly schedule.

In this issue, I cover:

• Notable News and Developments: AI startups losing founders, pushback on California’s AI Bill, NIST’s updated AI risk framework, Grok’s risky experiment

• AI Use Cases in Production: Blackrock, Salesforce, Harvey

• Directors’ Corner: AI Adoption among Asset Managers

• AI Coding Assistants Deliver High Returns

• A few AI Tools to Try

Enjoy. If interested, you can read past issues through this link.

---

Notable News and Developments

1. More high-profile consumer AI startups lost their founders to big tech

It might be easy for high-profile AI founders to get sometimes hundreds of millions in VC funding, but they are also prone to being poached by big tech these days, leaving their early investors and employees behind. It is a very strange phenomenon and huge challenge for investors, but also a reflection of the mounting scaling challenges of paying for hefty costs to train a cutting-edge AI model without the ability to recoup through pricing and retention in the consumer market.

After the Inflection founders went to Microsoft, Character.AI co-founders Noam Shazeer and Daniel De Freitas recently joined Google. Character.AI’s investors reportedly will receive $2.5bn, vs the $1bn valuation in March 2023. In a similar move, founders of AI agent maker Adept went to join Amazon. Such is great financial outcome for the founders but frustrating for investors who depends on 100x returns on their stars in the venture portfolios.

Readers of this newsletter will not be surprised by the unfair advantages of big tech vs startups in general-use and consumer-facing AI models. However, it gets us even more down the path of concentrated power in AI and leads to more hesitation in investments towards AI startups.

2. Pushback on California’s Senate Bill 1047 on AI

California's Senate Bill 1047 has faced significant opposition from both the tech industry and academic figures. The bill, introduced by State Senator Scott Wiener, aims to impose safety standards on AI systems with development costs exceeding $100 million. It mandates pre-deployment safety testing and includes provisions for shutting down AI systems that pose significant threats.

Critics, such as OpenAI's Chief Strategy Officer Jason Kwon, argue that the bill could stifle innovation and drive talent out of California, advocating instead for federal regulation to avoid a patchwork of state laws. Dr. Fei-Fei Li, a prominent AI researcher, has expressed concerns that the bill could impose unnecessary restrictions on AI developers, potentially stifling innovation and penalizing the open-source community. She argues that the bill could place undue liability on AI coders, which might dampen academic research and development efforts. This sentiment is echoed by other major tech companies and venture capitalists.

3. NIST Released Updated AI Risk Management Framework

The National Institute of Standards and Technology (NIST) has unveiled a new GenAI Profile within its AI Risk Management Framework, tailored for companies deploying generative AI. This framework focuses on four key areas: governance, mapping, measuring, and managing risks. It addresses unique challenges posed by GenAI, such as data privacy and environmental impacts, helping organizations integrate trustworthiness and align with legal standards. By adopting this structured approach, businesses can responsibly harness the power of GenAI while effectively mitigating potential risks.

4. Grok, X’s AI Chatbot, Shows Some Dark Possibilities of AI Without Guardrails

Elon Musk calls X’s AI Chatbot Grok the “most fun AI in the world” as it has very little content moderation. Within days of its image generation function launch, many bizarre Grok-generated images swept the social platform X, including violent, offensive and sexually suggestive content. Since then, some moderation rules seem to be introduced into Grok.

---

AI Use Cases in Real Production

A common complaint in the “AI hype” is that many AI experiments never reach the production stage to deliver real benefits to businesses. Here are some leaders that have found successes, explained in their own words. I hope this brings some optimism:

1. How AI is Transforming Investing | BlackRock

AI is revolutionizing investing by enhancing decision-making through data analysis and predictive modeling. BlackRock has leveraged AI and machine learning for several years to help deploy investment intuition at scale, as explained in this article. The technology enables more efficient risk assessment and identification of investment opportunities, fundamentally transforming traditional investment approaches.

“Compared to general purpose chatbots, the large language models (LLMs) that we use for security analysis are trained and fine-tuned on more narrow, curated datasets to perform specific investment tasks with a high degree of accuracy.” - Blackrock Systematic Team

2. Salesforce Unveils Autonomous Agents for Sales Teams

Even though still early in technology involvement, the deployment of autonomous agents is gaining traction in roles characterized by high volume of repeat tasks, simple context dependency, and non-deterministic outcomes.

Salesforce recently introduced two such agents: the Einstein Sales Development Rep (SDR) and the Einstein Sales Coach, designed to enhance sales processes internally, as detailed in this article. These roles benefit from agents due to their ability to handle diverse tasks and scenarios. For instance:

“Everybody needs more tools to qualify the pipeline and engage with the right customers at the right time. What we are doing with Einstein SDR is launching an autonomous agent that is going to help you qualify the pipeline, answer the questions of your prospects, give them contextually relevant, personalized engagement, and even schedule a meeting, handing it off to you as a human seller.” - Katan Karkhanis, EVP and GM for Sales Cloud at Salesforce

3. Harvey Observes Real Progress in AI Utilization among Legal Sector Clients

In this blog post, the leading legal AI startup Harvey shares that the utilization among its 100+ clients have risen from 33% a year ago to close to 70% currently. The adoption within these client bases are growing at a sustained to accelerated pace, indicating growing trust in AI generated outputs for critical, high-stakes tasks.

---

Board Directors’ Corner: AI Adoption Among Asset Managers

A recent KPMG semiannual survey of US asset managers revealed that 73% of the 120 respondents plan to start adopting AI within the next 18 months. However, the primary concern preventing them from embracing AI is the risk associated with data integrity, statistical validity, and model accuracy.

Source: KPMG Survey

While these are valid concerns, they should not discourage boardroom discussions about AI adoption. AI is an exponential technology, and board directors must adopt an adaptive mindset to stay ahead. It's crucial to develop key metrics for monitoring AI progress in your specific use cases.

For example, if you identified a few months ago that data integrity and security as the risk preventing you from AI adoption, you might be surprised by recent developments. Many cloud providers and enterprise AI applications now prioritize data integrity, security and privacy in their solutions. Some even offer services to help clients fine-tune and host their AI models privately.

---

AI Coding Assistants Deliver High Returns

The landscape of AI business applications has been fraught with challenges, particularly in demonstrating clear returns on investment (ROI). While many anticipated that AI would revolutionize customer service and marketing, the reality has been more complex, with issues like data quality and context-aware reasoning affecting outcomes. However, AI coding assistants have emerged as a standout success, offering compelling ROI and transforming the software engineering field.

The rise of coding assistants

AI coding assistants have gained traction due to their ability to enhance developer productivity significantly. These tools are particularly appealing in the tech industry, where software engineers represent a substantial cost. According to Financial Times, startups in this space have raised nearly $1 billion, with continued interest from investors.

Leading players include Replit, Anysphere, Magic, Augment, Supermaven, and Poolside AI, each contributing to the rapidly growing adoption of AI coding tools. GitHub Copilot, developed by Microsoft-owned GitHub in collaboration with OpenAI, exemplifies the early success of AI coding assistants. With nearly 2 million paying subscribers and more than 77,000 organizations as customers, Github Copilot has generated close to $1 billion in annualized revenue. This success story underscores the significant impact AI coding assistants can have on productivity and efficiency.

Cursor: the new AI coding tool that takes the software development space by storm

Among the latest AI coding tools, Cursor stands out as a best-of-breed AI-powered code editor. Developed by Anysphere, Cursor is designed to streamline complex coding tasks, allowing developers to turn brief directives into functional code, automate refactoring, and manage large-scale code changes rapidly.

The tool has quickly gained popularity among developers, small businesses, and large enterprises, boasting over 30,000 customers globally. Cursor's development was spearheaded by a team of MIT students, and the tool has received significant backing, including a $60 million Series A funding round led by Andreessen Horowitz.

Cursor's unique approach to simplifying coding tasks and enhancing developer productivity positions it as a formidable competitor to established players like GitHub Copilot. Many say it improves their productivity significantly more than what GitHub Copilot could.

The broader impact on software development

The adoption of AI coding assistants is reshaping the software development landscape. AI coding assistants have emerged as a powerful tool in the software development arsenal, offering substantial productivity gains and a clear ROI. Software engineers are usually among the most highly compensated employees and their productivity gains directly benefit company profits.

McKinsey estimates that these tools can impact software engineering productivity by 20% to 45%. Amazon CEO Andy Jassy highlighted the transformative potential of AI coding assistants, noting that they have saved the company $260 million and 4,500 developer-years of work.

“[The use of AI coding assistants] has been a game changer.” - Andy Jassy, Amazon CEO

Despite the promising productivity gains, challenges remain. The accuracy of AI-generated code and the need for skilled human oversight are critical considerations. Developers must balance the benefits of AI assistance with the necessity of maintaining high code quality and security standards.

Another exciting impact is the democratization of software creation among people who never consider themself ‘technical’. Many demo videos on X and YouTube share examples how you can pair Anthropic’s Claude and Cursor to create commercial applications with English and some design sketches.

Why does it matter to you as a non-technical leader?

Even though many of us will never be in software development roles, understanding how fundamentally this field has already been transformed by AI has tremendous implications in profitability, strategic planning, resource allocation, and product roadmaps.

I am excited by the possibilities, especially when so few people have paid enough attention.

---

A Few AI Tools to Try

The AI tool I’d most like you to try ASAP is the combo of Claude and Cursor, as mentioned above. In addition, here are a few interesting ones.

Gamma: AI for Presentation Slides

Gamma is an innovative tool designed to enhance business presentations with dynamic visuals and interactive elements, making it ideal for founders and business leaders who need to communicate complex ideas effectively.

OpusClip: Clip Relevant Content from Videos Based on Text

You’ll love this one if you work in marketing or content creation. OpusClip can transforms long videos into engaging short clips based on your text prompt within seconds, perfect for marketers and creators aiming to boost engagement and streamline video production, or for parents who want to impress their children…

Perplexity Pro: Free for One Year with LinkedIn Premium

If you already pay for LinkedIn Premium subscription, you should have received an email offer of one year’s free subscription of Perplexity Pro, a service that costs $20/month if paid monthly. Definitely take advantage of this offer.

[BTW, Perplexity’s recent pricing and GTM strategy have been confusing and I may write something about this in the future. ]

---

One More Thing

On my long trip, I felt the impact of AI every day on my trip...but not always in the ways I'd like it to be. I really do not need more AI algorithms telling me what to buy, where to go, what to eat, and worse, how to think. Instead, I want AI to help offset the declining productivity, take care of our seniors, empower people to create, and bridge our differences. How can we get there?

I hope you found this newsletter valuable. If so, please consider sharing it with other leaders and board directors in your network. I greatly appreciate it.

Thank you.

Joyce Li