AI Agents // future of AI or just hype?

sbagency
5 min readOct 1, 2024

--

The hype around AI agents is constantly heating up // technically, agent technology is just an abstraction of multi-stage generation, including the use of tools, code generation and execution, memory, and, of course, calling LLMs.

AI Agents are able to imitate almost any cognitive function that humans do, not good right now, but sooner or later the quality will rise.

Here’s a summary of the key points from the conversation between Jensen Huang and Mark Benioff:

1. Transition in the tech industry: The industry is moving from being focused on tools (computers, software) to being centered around skills and AI agents.

2. AI agents’ capabilities:
— Understanding and reasoning
— Using tools
— Collaborating with each other
— Solving complex problems by working together

3. Technological advancements:
— Unsupervised learning as a breakthrough, allowing AI to expand capabilities beyond human labeling limitations
— Progress faster than Moore’s Law, potentially at “Moore’s Law squared”
— Advancements in CPUs, GPUs, and machine learning creating a rapid feedback loop for AI improvement

4. AI safety and development:
— Importance of fine-tuning, guard-railing, and safety measures
— Using AI to curate data and create safe curricula
— Implementing reflection and chain-of-thought reasoning in AI responses

5. Practical applications of AI:
— Making AI accessible and understandable for widespread use
— Simplifying the process of building AI agents
— Comparing onboarding AI agents to hiring employees rather than traditional software development

6. Historical context:
— Reflecting on the evolution from early personal computers to current AI systems
— Excitement about solving long-standing challenges with new AI capabilities

7. Enterprise AI era:
— Celebration of the beginning of enterprise AI
— Potential for unprecedented levels of automation in industries and society

Overall, the conversation highlights the transformative potential of AI, particularly in enterprise settings, while also addressing the need for responsible development and practical implementation.

https://aiagentsdirectory.com/landscape

The AI Agents Landscape Map offers a detailed overview of the current state of AI agents, categorizing them by functionality. It provides insights into the diverse capabilities of AI agents and their applications across various sectors. This resource is valuable for developers, business owners, and researchers interested in integrating or studying AI technologies. For the latest trends and advancements in AI.

https://www.felicis.com/insight/the-agent-economy

Generational platform shifts create new economies, and recent AI innovations signal the rise of the “agent economy.” Similar to how public cloud services gave birth to the SaaS economy and the iPhone spawned the app economy, AI agents are set to transform business landscapes.

Key reasons for investing in AI agents include:

- Reinvention of SaaS: AI agents disrupt traditional SaaS models by moving away from point-and-click interfaces and siloed data.
- Labor Budget Integration: With labor budgets significantly larger than software budgets, AI agents can tap into both.
- Service to Software Transition: AI agents enhance efficiency in low-margin human services.

The agent economy encompasses various applications, categorized into horizontal (industry-wide), vertical (industry-specific), and consumer-facing solutions. Examples include legal AI for in-house teams, law firms, and consumer services.

Tech giants are expected to invest over $1 trillion in the coming years, indicating a robust application ecosystem is on the horizon. Early winners in sectors like customer service, sales, marketing, and healthcare are emerging, but there’s still significant potential for growth. While the timing for major breakthroughs in AI applications is uncertain, early signs of success suggest that transformative AI agents are already in development, poised to impact various industries and roles.

https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/why-agents-are-the-next-frontier-of-generative-ai

Here’s a summary of the key points from the article:

1. Generative AI (gen AI) is evolving from knowledge-based tools to AI-enabled “agents” that can execute complex, multistep workflows across digital environments.

2. Gen AI agents can use natural language, plan actions, utilize online tools, collaborate with other agents and humans, and learn to improve their performance.

3. The technology is still nascent but attracting significant attention and investment from major tech companies and startups.

4. AI agents can potentially automate complex use cases that were previously difficult to address due to highly variable inputs and outputs.

5. Three key advantages of gen AI agents:
— Managing multiplicity (adapting to various scenarios)
— Using natural language for direction (easier for non-technical users)
— Working with existing software tools and platforms

6. The article outlines a four-step process for how gen AI-enabled agents could work:
— User provides instruction
— Agent system plans, allocates, and executes work
— Agent system iteratively improves output
— Agent executes action

7. Three potential use cases are explored:
— Loan underwriting
— Code documentation and modernization
— Online marketing campaign creation

8. To prepare for the age of agents, business leaders should consider:
— Codification of relevant knowledge
— Strategic tech planning
— Human-in-the-loop control mechanisms

9. While the technology shows promise, it’s still in early stages and poses various challenges and risks that need to be addressed before widespread deployment.

Here’s a summary of the key points from the interview with Jake Heller of Casetext:

- Casetext was founded over 10 years ago to improve legal technology. For the first 10 years, they made incremental improvements to legal workflows.

- When they got early access to GPT-4, they pivoted the entire 120-person company in 48 hours to build an AI legal assistant called Co-Counsel.

- Co-Counsel can perform tasks like reading millions of documents, doing legal research, and writing memos — work that would take lawyers days to complete manually.

- They used a test-driven development approach, writing thousands of tests to ensure 100% accuracy, which is critical for legal work.

- The product involves much more than just wrapping GPT — it includes proprietary datasets, integrations with legal systems, OCR capabilities, and carefully engineered prompts.

- They went from a $100 million valuation to a $650 million acquisition by Thomson Reuters in just 2 months after launching Co-Counsel.

- Jake built the initial prototype himself and convinced skeptical employees by showing early customer reactions.

- He believes there’s huge potential for AI to transform many industries beyond just legal, as long as companies focus on getting to 100% accuracy for mission-critical use cases.

- On GPT-4 vs Claude AI, Jake noted Claude’s ability to do more precise, detail-oriented thinking and analysis on complex texts.

The interview provides insights into how to successfully build and deploy vertical AI solutions in enterprise settings.

--

--

sbagency

Tech/biz consulting, analytics, research for founders, startups, corps and govs.