Automate Thoughtfully with Agents
Much of my time is spent aggregating information. The aggregation process can be summarized in a 3R framework of Receive, Review and Return:
- Receive information [inputs; the information to aggregate]
- Review and do work with the information [processes; the process for aggregation]
- Return insights and impact [outputs; the results and impact of aggregation]
The aggregation process tends to be automated when:
- The set of tasks are many and straightforward
- There is a net upside to automation based on returns relative to required resources
For many folks, the “aha! moment” using agents may come with the automation of process, and this may be because the perceived trade is “quickly done and well done work” for “a marginal amount of money, more time and more possibilities to deliver a positive impact”.
On the point of “marginal amount of money”, enterprises may need to approach automation more thoughtfully with agents. For any work that an agent does the 3R framework may be applied as follows:
- Receive information -> Token inputs are received
- Review and do work with the information -> Tokens are processed
- Returns insight and impact -> Token outputs are returned
Token prices¹ vary based on which R in the 3R framework an agent is working on. For example, refer to Anthropic’s Claude Team & Enterprise pricing here and Alphabet’s Google’s Gemini for Google Workplace pricing here. Given that the prices are variable, the usage of these tools by employees and entrepreneurs put at risk the notion of trading “quickly done and well done work” for “a marginal amount of money” and other upsides. Furthermore, there is still the risk that the other notion that agents complete “well done work” introduces the need for additional technical operational and other non-financial resources in the future to improve, repair or replace “not well done work”
Before the “agents era”, processes like aggregation processes were already automatable. Therefore, using agents to automate an aggregation process can be most helpful when asking the agent to build automated systems involving people, processes and tools. In this way, the agent is delivering work to enable employees and entrepreneurs with an upper bound on financial cost, as compared to the agent is doing work without an upper bound on financial cost.
To conclude, prioritize using agents as architects over using agents as executors.