In my previous post in this series, I shared some context behind how I ventured out to start building an AI job search agent and journeyed into the land of crystallizing the difference between an app and an agent. This post provides another window into that journey.

The Presentation Layer

I introduced the concept of an ARD (Agent Requirements Document/Documentation) - essentially a PRD but for AI agents. As I was writing the ARD for the job search agent, I started thinking about the details of the presentation layer. Given the need for user inputs (e.g. resume details and role preferences) and the need for a human-readable output (e.g. for suggested roles and companies), the design started shaping up much like an app fairly quickly.

From my understanding of what AI Agents are, there are versions of AI agents that “can just do things” for the user. Given the right tools, access, and permissions, the agent can act on a user’s behalf and sometimes the output is a completed process. In the case of job searching, the output might be navigating all the way to the point of submitting an application - already filling in pertinent information and attaching the user’s resume. That sounds interesting - and has its benefits and challenges.

However, I realized that what I set out to build was more of a discovery tool. Imagine: “Help me discover something” vs. “Help me do something”. In other words: “Show me something” [the app use case] vs. “Do something” [the agentic use case]. Sure, an AI agent could still “help me discover something” but the UI layer is still pertinent - for presentation purposes and also for interactivity.

Manual Pieces of An Otherwise Automated Flow

A job search is not necessarily a “set it and forget it” type of process in the way making an appointment or booking travel might be. Users will likely want to adjust preferences over time and will likely have updated resumes as time progresses (at least until their next great opportunity is secured!). The general analysis, search, and presentation process can run in an agentic fashion but making updates to the criteria used by that process still remains manual.

Additionally, I wanted the user input interface and output surface to be presented side by side and I added details around this in the ARD. I think a typical agentic approach could work by presenting the results in a list or even by generating a new UI with each output (infra considerations aside). However, given the desired user experience and presentation modality, it became more and more apparent that an app was the better option for the time being.

In summary, when considering what my job search tool would look like, it being an app rather than an agent became the brighter option.

The Series Continues

The story isn’t finished for now - I plan on delving into what an AI agent is in another post.

Posting soon,
Jewel

–

Extra Tidbits:

In case you’re wondering about the status of the job search tool discussed above, I stopped work on it, for now, to focus on other things. 

If you’re currently job searching, feel free to leverage the process below.

1. Gather resume details that you want analyzed

• Be mindful of sensitive information. You don’t have to upload or copy/paste all of your resume details into an AI tool for this to be beneficial to your search.

2. Use an LLM to analyze the details

• Try a prompt like this one - “Please analyze the following resume details and list suggested roles including currently open roles and links to job descriptions” followed by the resume details.

3. Iterate on the prompt if needed until you get a list of open roles with links. You might learn about a company you haven’t heard about before!

• You can also iterate on the prompt by mentioning specific job titles to help steer the results in a direction helpful to you.

4. [Recommended] Keep track of companies that you’ve checked and the last date you checked their careers site for opportunities that might be a fit for your background and what you’re looking to do.

• This can help you approach your job search in a methodical, organized fashion.

If you too are a Believer in Jesus Christ, remember that we can reverently pray about everything - including a job search - and ask Our Father for what we need in Jesus' Holy Name. I recommend that you pray over your job search. That will be better than any and all other processes I can recommend to you. Pray first (wisdom). Pray in the midst of the search (encouragement). Pray once you have your new great opportunity (thanksgiving). Keep praying.

Scripture for Encouragement: 1 John 5:14; Luke 11:9-10

Happy New Year!

2024 is here and I'm so grateful to have been able to spend time reflecting and looking forward. Something I've wanted to return to for some time is sharing some of the ideas, thoughts, and learnings I've gained as I've delved into new areas in my career. I have blogged before and I even started a podcast at one point. In recent months, my desire to 'return' to this type of sharing in these professional areas of my life has really grown. In considering the how, what, when, and where, I've returned here to this blog! The structure, name, location, etc., etc. of this blog may change but I'm starting this new chapter, here and now.

Stay Tuned
I have a number of post ideas planned out covering topics in tech and career journeying.

Along these lines, do you have any questions or topics that you'd like to see me address? Let me know!

Happy Tuesday and thanks for reading!

Until the next post,
Jewel

As you may or may not know, I have worked in the advertising technology industry (ad tech for short) for about the past two years. In April, I attended Programmatic I/O – a major industry conference - for the first time. It was stocked full of amazing content and truly amazing food.
 
I’ve developed a Recap that covers a bit of what I learned over two extremely educational days.

For the sake of digestibility, I’ve broken up my Recap into three parts. The first part covered new terms and tech I was introduced to, the second part covered major trends I noticed, and the third part part below covers session content.
​
Without further ado, here are some one-liners from the Programmatic I/O sessions I was able to attend in SF this year!

Session One-Liners:

• Programmatic: Key Numbers, Key Trends
  1. ​​​"What is Programmatic? - An automated method of buying, selling, and fulfilling digital ads."
  2. "Though the shift in share of spend is leaning to mobile, eMarketer found that desktop still commanded a higher CPM."
  3. "The audience  focus is important in programmatic."

 

• How to Use Transparent Data to Reduce Programmatic Waste
  1. ​​"Data is like water: it's abundant and badly needed."

 

• Programmatic In-House: Best Practices in Agency/Marketer Relationships
  1. ​​There are 3 "big things needed" to transform programmatic operations in-house: "Tech, Targets, and Talent".

 

• A Market Perspective on Apps
  1. "Gaming is dominating app monetization"

 

• Best Practices for Implementing a CDP
  1. ​​The "Aggregation Tier" is where CDPs (aka customer data platforms) live in the AdTech stack.
  2. "In a CDP, you need to be able to control what data goes where."
  3. After designing the CDP, map the relevant services -> "Collect, Control, Connect"

 

• Mobile Fraud
  1. ​​"If there's no advertising, there's no point in paying for results since ads didn't drive the results."
  2. When looking into detecting fraud, look at your partners (spend v impressions, spend v app installs, etc.) and ask: "How are these partners paying for their supply? Does the story you're hearing sync with your intuition/experience/industry norms/industry standards/etc.?"​​

 

• How to Build & Leverage A Unified Data Stack
  1. ​​​​"Think of your data stack as the body's nervous system & everyone in the company is a customer of it."
  2. "Think of the 'getting of the data' separately from the 'processing/using of the data'".
  3. "Collect more than you need, less than everything -> back-up everything."

 

• How to Get More Out of Your Data
  1. ​​"Data can help you develop what not to do, as well as what to do."

Plenary Program One-Liners:​

• Shades of In-Housing: ​​"Perhaps the best in-house move is a psuedo-in-house move."
• Activating the Brand: "Focus on both in-house & partner expertise -- what brands want are trusted advisors."
• Digitization of TV: New term - GRP
• Omnichannel Buying Remains More Promise Than Reality: (Joanna O'Connell) "Reaching me is important but reaching me in this 'moment' is even better."

Want to learn more about programmatic advertising? Great! The East Coast version of Programmatic I/O is coming up soon - October 15-16, 2018! Learn more here: Programmatic I/O NYC.

Next Up: AWS' NYC Summit Recap

Do you have any recommendations for other conferences I should attend? Share them with me! :D