In a previous blogpost, I wrote:

Since publishing that blogpost, an acquaintance posted this about his property management company’s¹ use of agents to build software to maximize information consistency across its online, active property listings. My acquaintance wrote that, “The whole thing runs itself (without using tokens every run).” Based on this, it appears that the post shares one example of an agent which successfully created software to automate one of the PMC’s systems involving manual processes executed by people, processes and tools.

After reading the post, one may wonder, “How does one successfully create software to automate a manual process?” To achieve this, there may be 4 points for consideration.

Firstly, one needs at least one measurable result. The measurable result serves as a means to validate that an action or set of actions causes the desired outcome or set of outcomes. In the case of the PMC, there may be multiple measurable results which include and are not limited to a revenue-benefit-focused metric such as “time-to-rent”, an operation-cost-focused metric such as “time-on-curation” or a coordination-cost-focused metric as “time-on-vendor-management”.

Secondly, one must articulate the pain points which prevent the best possible results. The pain points serve as means to validate that there is at least one value-creating problem to solve. In the case of the PMC, one pain point may be the different costs associated with managing its property unit listings across its portfolio of property communities. To illustrate, the PMC has N1 property communities x N2 property unit listings x N3 property management platforms, and each element unit listing will require T time spent on curation per unit listing. As a result, the PMC’s pain point is the total time spent on curation which can be formulated as T x N1 x N2 x N3. With an understandably large total time spent, PMC employees may be overwhelmed by the task, and may be discouraged to complete such a task given the risks for error considering the dataset size. Consequently, the task being not done well or the task not done at all could result in costs such as the financial-cost-focused metric of “revenue loss” or the reputation-cost-focused metric of “negative feedback”.

Thirdly, one requires a comprehensive, effective and scalable manual process. The process serves as the foundation for building an automated system. In the case of the PMC, the team appear to be subject matter experts in online, active property listing management. As such, one may assume that a comprehensive and effective process exists. On the point of scalability, this blogpost defines scaleability as the ability for people-in-the-process-loop to be fungible. Put another way, if the same process may be passed from P1 person to P2 person, then the process is scalable. Based on the multiple people cited in my acquaintance’s post, the process appears to be scalable.

Fourthly, one must develop a basic understanding of internet information transfer. This blogpost defines internet information transfer as the information that is accessible in an internet browser session. Here are some factors for consideration:

Thus far, this blogpost opined that one may successfully automate with agents when 4 points of consideration are planned for:

Thinking through the 4 points above, one may wonder, “To automate successfully with agents, there appears to be a lot of work”. In response, this blogpost holds the view that there is indeed a lot of work required, and that work results in the formulation of a specification document as a prompt. On 2026 July 20, Cursor advanced this idea of “specs as prompts” in their blogpost here.

To conclude, automating successfully with agents is a lot of work. Therefore, choosing what results to drive, and what pain points to solve remain important. If one wonders, “How do I prioritize those results to drive and/or pain points to solve?”, then consider reading this previous blogpost of mine Ranking Is Important.