AI and the Agency Opportunity
Agencies Built From the Ground up for the AI Era with AI Driven Operational Efficiency and Startup Scalability Through Productization. Focused on Value Add through Creativity and Imagination to Leverage New Possibilities.
Intro
AI represents a transformational change, different jobs and skills will be in demand, different types of investment will be needed and different types of organizations will flourish.
Taking all the above changes into consideration, below, we argue that organizations more like agencies rather than traditional startups are best positioned to capture a large opportunity that has recently emerged.
Bullet point summary:
- AI will transform the way large businesses operate.
- Agency like companies, not traditional startups, are best able to capture the end user, interface side of the barbell opportunity.
- Capturing the interface opportunity, prioritizes ideas and creativity, commoditizing technical execution and shifting focus from distribution to production.
- Corporates increasingly require customized AI solutions based on proprietary data, presenting a substantial opportunity for agencies serving them.
- The AI service spectrum includes commoditized back-end systems, middleware for prompt customization, and innovative front-end interfaces.
- Agencies with creative/strategic talent are essential in the AI era, focusing on generating unique AI prompts and innovative use cases, which can be productized.
Market Opportunity
A business consists of: Ideas > Build > Distribution > Sales.
For early stage startups, the order of importance of these in terms of differentiation are: 1. Distribution 2. Build 3. Sales 4. Ideas
However, in the AI era, ideas become more important, sales & distribution are in lockstep, much more of the technical execution is commoditized and what we traditionally call agencies are better suited to serve it, by getting to revenue quickly through creative value-add.
The current model of venture backed startup investment exists because of the distribution advantages that the internet created (network effects and virality). These meant that some, winner-takes-all services could grow very large, very quickly and in advance of revenue, where lots of others would fail. Only equity financing would give the shared upside in the winners to offset the high risk of failure and startups needed financing to fuel the gap between growth and revenue as they could not bootstrap with sales.
Agencies did not fit in this model as they did not scale as quickly, whereas ideas weren’t needed in volume in startups (most successful startups are well executed, better mousetraps, using commoditized, open-source technology but with unique distribution from network advantages, not unique ideas).
AI changes this. It makes software itself a network and, as as argued here, the network effects will play out in production rather than distribution. Software and things produced by software will become cheaper and more plentiful and the scarcity will be in the compute, data and in the ideas for things we can get ubiquitous, low cost, custom software to do — more broadly, the questions to ask the machines that respond to them.
The compute power will be provided by large platforms who will also deliver models running on their own or public data. Large corporates will augment that with their own datasets. Every corporate will have models with their own, proprietary information augmenting foundation models, since siloed, proprietary data will be a source of increased advantage of paramount value.
This will mean that every corporate is a potential customer for custom AI use cases and will devote some budget towards it, creating opportunities for B2B businesses with connections to corporates. In some ways this is no different from the early days of the web and the emergence of website and/or intranet building agencies, except that the lower level of creativity required there and modest level of execution requirement made this a fragmented market serviced by largely undifferentiated suppliers with little productization. The AI opportunity on-the-other hand will require more unique differentiation through unique ideas and so is less likely to commoditize the way that web design did. The opportunity for AI era agencies is much bigger than for predecessors such as Razorfish.
In summary, the AI era presents a barbell opportunity for businesses, with giants from chip makers, cloud providers and platforms on one side and the important thinner layer of end user use cases and interfaces on the other which requires imaginative skillsets, to deliver on.
The portion in the middle, the traditional sweet spot of venture backed custom software products is being commoditized by the larger players. The remaining interface layer needs to be serviced by sufficiently creative people, not sufficiently robust software development execution and so the time it takes to deliver value is not tied to production volume and can have high margins. In addition, since the route to corporates is served by their universal need to run over their proprietary information it is possible to bootstrap these kinds of creative business and productize the offering for recurring revenue. For this reason, consulting and agencies may be sexy again as they emerge from the worst period in more than a decade.
Product Opportunity
Every corporation needs AI capabilities based on their own internal data and processes to leverage increased process efficiency and data advantage.
Delivering on this opportunity consists of three layers:
Back End
The cost of the back end for a custom AI system has dropped dramatically and is being commoditized. Going forward, most corporates will choose to build on top of an existing LLM and cloud hosting service. The opportunity for this is rather like the early web and production of Intranets, which eventually became part of B2B SaaS services and there is a lot to learn from the successes and failures here. Building custom models from scratch is not needed for most applications and neither is fine tuning. Making internal data part of the context of a prompt is usually sufficient.
Middleware
Middleware can exist to create systems that are agnostic in terms of what LLM and cloud service to send a prompt to. This enables those creating prompts and interfaces to both create a switching cost and to service at scale with repeat offerings.
Front End
Although LLMs represent a new form of interaction with computers through natural language, where new capabilities such as truly no-code software development exist, they still have all of the problems traditionally associated with command line interfaces.
Command based interfaces have a design issue, you need to know the commands. In the language of the famous interface designer, Don Norman, they were not ‘discoverable’. You couldn’t mess around and discover the capabilities of a system in the way that you could with a mouse and icons interface. AI implementations need the equivalent, they need the prompts and custom LLMs to be represented by buttons and avatars that people can easily discover and use. They need new interfaces.
But the problem goes further, as anyone who has owned an Alexa device which is used as a glorified egg timer or voice activated radio, knows. It’s not just an interface problem, nobody really knows what the capabilities are — what questions you can ask of a natural language AI system.
Asking the right questions has been a cornerstone of philosophical enquiry from Socrates to Einstein. Coming up with great questions is a task that favors creativity over engineering skills and this is a massive opportunity for consulting organizations with creative or strategic talent.
Much of the current focus at the interface and of AI has been focused on the ‘dark art’ (more like alchemy than science) of prompt engineering. But browsing through prompt marketplaces such as Promptbase and it becomes immediately apparent that the problem isn’t just prompt engineering it’s prompt imagineering — coming up with ideas for prompts.
Coming up with questions, crafting them into prompts, managing them as interfaces and menus and running them over LLMs enhanced with proprietary data is the single biggest service opportunity in a generation, for business services.
Creative Opportunity
As AI lowers software build costs, there will be more use cases that are cost effective to build and in some cases these will just be custom settings on foundation models, menus, buttons and prompts — what we are calling ‘questions’. Highly creative people who can come up with the questions to ask to unlock the potential of these new systems and build services around these will be in demand and these people tend to operate within creative agencies rather than startups.
To highlight exactly what we mean by creatives, we need to distinguish between craft and creativity, the camera highlighted this distinction 150 years ago but AI will apply it to all media and knowledge production.
By creativity we do not mean most of the people who are hired as ‘creatives’ based on, say, graphic design skills. We mean people who can think of new things and think differently. This same divergence happened on a less dramatic scale with traditional software automation such as the development of the spreadsheet, where the number of bookkeepers rapidly declined but accountants increased [graphic]. Accounting required higher level ‘creative skills’ than bookkeeping (even if creative and accounting are two words that should never go together).
When you strip away all the craft of creative production, which AI systems do, all you have left is the instruction sent to the machines that execute the craft — the questions. Amazon’s Alexa is an AI system that has been around for nearly a decade and it still has the same problem in the most literal sense: people don’t know what questions to ask. With ubiquitous AI, people will need to know what ‘questions’ to ask (i.e. creative ideas of things it can do) and corporates won’t have the time to just mess around and find out.
The ‘questions’ won’t literally just be a bunch of text, we mean questions in the abstract sense i.e. use cases, personas, menus and they will be wrapped up in everything from business strategy, applications and processes.
What specifically will these services look like and how will they be productized, to avoid the agency trap of revenue as a function of people?
They will consist of business strategy, business design, software, projects, campaigns & training and they will be delivered via different layers of interface productization:
- Questions — browse around the biggest prompt marketplaces and you will see that prompt engineering, although a dark art is less important than creative questions. Most prompts fall in to a relatively narrow range of use cases.
2. Styles — as a non deterministic technology, AI is very good at the qualitative over the quantitative — styles vs workflows. This is a shift away from how corporates tend to view computing capability.
3. Personas — the culmination of style is where the style attaches to the entire personality of a chat interface i.e. building custom personas. This is something where, as usual, the counter culture areas of technology have innovated first, as per the below screenshot. Delivering personas rather than workflow tools for a corporation allows for open ended rather than specific capability, like having an army of interns that can be activated on demand.
4. Menus & UI— not only do AI systems need interfaces and menus, they will be completely different from the workflow based ones that have existed until now. They will be able to accomplish complex processes in one, command that embodies the meaning of the desired action. I am calling these ‘semantic interfaces’. Delivering semantic interfaces will be cost effective on a highly customized level, for corporates.
Conclusion
Businesses that are more like agencies than traditional startups, are uniquely positioned to exploit emerging market opportunities by capitalizing on AI-driven operational efficiency and scalability to serve corporate clients at scale, who all need system to run over their proprietary data and processes.
As AI lowers the cost of software development and intensifies the need for creative problem-solving, certain types of agencies will emerge as crucial players, adept at navigating this new landscape. They will offer tailored AI solutions, leveraging creative and strategic insights to meet the evolving demands of corporates. The ability to generate innovative ideas and craft effective AI prompts, use cases and interfaces becomes a key differentiator, positioning agencies who can deliver at the forefront of this transformative wave.
