By
Executive summary:
Key take-away: Companies should be building AI Agents now with a developer team & business leader to understand the implications they bring to how we will work going forward.
AI Agents will be like excel spreadsheets. Anyone will be able to make them.
Prompting - telling AI what to do - will be as important as entering a formula into the cell of a spreadsheet.
The more advanced one is with prompting the more advanced the AI Agent one can make. However, unlike spreadsheets the AI Agents will be able to 1) converse with humans and 2) work with other agents.
The way we work will change, new business models, pricing models and org structures will be developed.
TLDR:
- Every white collar worker will have an AI Copilot, then an AI Agent.
- Creating & using AI agents will be as common as using a spreadsheet
- Individual's new workflow will be: To think with AI then work with colleagues
- New Business models: shift from Software-as-a-Service to Service-as-a-Software
- Pricing: Usage based pricing will become more common impacting Sales Rep. compensation and revenue forecasting
- Links to 6 tools to build AI Agents
- The AI economy formula: AI ROI = (Business Process Value x Volume) / AI Costs
In August, 2011 Marc Andreessen published an article in the Wall Street Journal with the title “Why software is eating the world” in which he shared his theory that “we are in the middle of a dramatic and broad technological and economic shift in which software companies are poised to take over large swathes of the economy.”
He cited the following examples:
- Amazon surpassed borders as the largest bookseller
- Netflix surpassing Blockbuster as the largest video service
- Apple iTunes as the dominant company in the music industry
- LinkedIn as the dominant company in the recruiting industry
- Pixar as the best movie production company
Well, 13 years and many articles later, Marc Andreessen, in a new article, recently said,
“Every white collar worker will have an AI Copilot,
then an AI Agent.”
Let's put some definitions to these terms before we go on.
Definitions:
- An AI Copilot is Microsoft's product name for its AI Assistant. The pilot or lead in is the human and the copilot or assistant in the relationship is an AI helper that supports the human with a focused task often laid out in steps through a series of internal prompts.
- An AI Agent is similar to an AI assistant except it has system instructions that enable it to operate autonomously with a set of tools at its disposal to operate with in its given role and instructions.
- Difference: An agent decides what tool to use when and how instead of having to be specifically told step by step instructions.
Back to Marc Andreessen's recent quote, The idea that every "white collar" worker will essentially have an AI assistant isn't a farfetched one. Most people already use an AI Assistant (Siri, Alexa, Google) for simple daily activities.
The release of ChatGPT though brought an ai assistant that could code and accomplish complex tasks putting it and other LLM's on the path to be an AI work assistant.
Macro Impact:
Scale matters.
Usually incumbents are slow to move during technology disruption, but in the case of AI disruption scale matters for incumbents.
The chart below shows how larger software companies have had greater stock returns to investors since the release of ChatGPT compared to their smaller peers. This isn't just about the hype cycle its also about AI strategy and speed to market.
The majority of the 16 companies in the >$20B EV cohort have released their own AI Assistant or in the case of Salesforce (CRM), Hubspot (HUBS) and Palantir (PLTR) they released an AI Assistant platform, where as the smaller cohorts have mainly integrated AI into their product.
- Databricks (DBX) is an outlier in the smaller cohorts as it has built an AI platform thanks to its acquisition of MosaicML, a LLM company.
Slide credit: https://www.platformaeronaut.com/p/enterprise-software-ai-pricing-power
Thomas Reiner author of Platform Aeronaut argues that "the benefits of scale, are an entrenched customer base, and most importantly a ton of unique data stands to argue that the larger players will continue to benefit."
I actually think the advantage companies at scale have is their capacity specifically in terms of cashflow operations and talent.
Companies at scale:
- Can select a talented team to focus on AI initiatives through engineering, product development and M&A.
Companies not at scale:
- are adjusting from "growth at all costs" to "efficient revenue growth"
- are limited in their operational and talent capacity
- resulting in AI Initiatives often becoming side projects for an individual or team who is already often or near capacity.
- AI initiatives get pushed off until other priorities are addressed
Interested in learning frameworks to achieve efficient revenue growth? Click here
Scale enables second mover advantage
The disruptors, OpenAI & Anthropic, clearly achieved first mover advantage as they are now valued at $154bn and potentially $40bn respectively. Pretty good for two companies no one new of two years ago.
The incumbents that were already at scale and able to to start shortly after the launch of ChatPGT got a least a twelve month head start of incumbents that were delayed at least six months.
What? Why 12 months?
The new law of the jungle is the law of AI scale which states that AI capacity doubles every six months. So the first movers essentially got a 12 month advantage over the fast followers and a 24 months advantage over the early adopters.
"AI capacity doubles every six months"
- the Law of AI Scale
Speed of AI multiplies the advantage
With capacity doubling every six months, a company that hasn't started building AI assistants, or AI Agents today is not 24 months behind but rather 48 months behind due to the amount a development team has to learn and grasp the possibilities. Companies that don't get started in the next six months will be 96 months (8 years) behind the first movers.
Luckily, there is hope!
The cloud giants - AWS, Microsoft, Google - have released tools to accelerate the development of enterprise grade AI Assistants and AI Agents. These no-code Agent builder tools simplify (not solve) the AI Agent development workflow.
Rise of AI Agent Tools
AI Agent Tool links:
- Google's Vertex AI Agent Builder
- Salesforce Agent force
- Microsoft Co-pilot studio
- AWS Bedrock
- ServiceNow AI Agent platform
- N8N no code canvas to build agents
September 2024 Marc Benioff talked about how AI Agents will change the workplace.
A month later, Satya Nadella gave a more clear analogy to how AI Agents will change the workplace by comparing them with excel spreadsheet. He believes working with and creating AI Agents will be as natural as creating and using an excel spreadsheet.
“Just like how you could create an Excel spreadsheet that was a forecast,
you can now create AI agents.”
- Satya Nadella, CEO of Microsoft
What does it mean going forward if creating an AI Agent will become as easy as using a spreadsheet
Need help creating AI Agents? Click here
The way we work is going to change.
Business models will be disrupted
- Similar to how the internet enabled the Software-as-a-Service (SaaS) business model which changed how software used to be sold from in stores on a CD to being purchased online via a subscription.
- Service will come from software creating a Service-as-a-Software business model
credit: https://foundationcapital.com/ai-service-as-software/
Pricing models will change
- Usage based pricing (UBP) or variants of UBP will replace or be added on top of subscription pricing
New Business metrics will be created
- Monthly Recurring Revenue (MRR) will evolve into just Monthly Revenue with subcategories for subscription revenue and usage revenue
- Finance & Analytics teams used to reporting MRR will have to break down Monthly Revenue by customer count into usage buckets of High, Medium and Low usage.
Sales Compensation will evolve
- Sales reps will shift from a focus on customer acquisition to customer acquisition + usage. The roles of sales reps and customer success reps will blend together.
credit: https://www.growthunhinged.com/p/from-selling-access-to-selling-work
Micro Impact:
The New Work Flow:
Satya Nadella, later in his keynote, went on to explain his new workflow, “I think with AI, and work with my colleagues at work” and went as far as to say it will become the new workflow for knowledge workers.
“I think with AI, and work with my colleagues at work”
- the new workflow according to Satya Nadella
New workflow comments are at the 8:00 min mark
Fundamentals for success
Understanding the prompt, prompt sequencing, and agent workflow will separate basic, power and advanced users just as writing an excel formula, building a multi-tab model and connecting to an external database separate basic, power and advanced Excel users.
A popular youtube developer, IndyDevDan, declared,
“The prompt is the fundamental unit of knowledge work.”
Fundamental unit of knowledge work: 4:46 -5:16
If the last twenty years, knowledge work was about the ability to understand how to use a cell in Excel or influence the numbers that get reported in an Excel cell then the next evolution is understanding how to use prompts to build and work with AI Agents.
Formula for Success:
AI ROI = (Business Process Value x Volume) / AI Costs
Before you run off and get your company to tackle a high value business process or a low value process that has high volume I recommend a crawl, walk, run approach just to learn the workflow needed.
- Build a proof of concept - the generative AI is achieving the task its supposed to do
- Push it to production - a user facing domain or an integrated chat user in slack or integrate into an existing automation workflow, etc.
- Publish MVP - provide documentation, support, testing & mgmt. oversight, iterate on user feedback
Once your development team has figured out the workflow & integrations it will be able to rapidly expand and improve its AI development capability and your AI ROI.
Remember, new business processes will emerge as the way we work is going to change.
In my next post i'll lay out the AI Agent Ecosystem. Don't worry it will be a shorter blog post.
Schedule some time with me if you found this interesting? Click here