From Agenda to Agent: Building a Decision Engine for Conferences

Over the years, I’ve realized something simple: conference attendance is a constrained optimization problem:

The buzzword density and overhype turns conferences into cognitive overload machines. Remember a normal brain burns about 300-400 Kcal per day at rest. Here it feels like 800 at a conference! Being an avid cyclist and ex-marathoner, I am very cognizant about limitations on physical energy as well.  Conference sessions are energy and time traps, imagine you want to run a race through Disneyland! You overbook sessions in halls far from each other, and then when you finally get there and find a seat, you see the content is just nearly over your leave-threshold.  Panels can quickly drift into promotional language and unfamiliar jargon or worst: redundant outdated material. 

On the other hand, most people are intellectually curious about latest technology and business trends to varying degrees. As a result, despite the limited time and unclear ROI, most attendees optimize for FOMO and still browse and plan agendas emotionally.

So how do you plan your conference attendance balancing curiosity, ROI, and physical/ mental capacity, so your brain and body are not too fried by 3pm to benefit from the most valuable conversations? May be a decision engine that takes the emotions out of it would help. 

I have become an workflow geek ever since I started an AI pay service. If something is repeatable, it should be systematized. My mind goes: “Build a workflow”! So why not a decision engine for conference attendance?

Somebody asked me recently can you send your financial statement handling workflow to me and I thought about how can a workflow that I build for this be explained or better yet: can be repeatably built by others. 

so here is how I started:  

Step 1: Define the Input Parameters

-My top technical focus areas and explicit exclusions
-My buzzword tolerance (10-minute rule)
-Panel tolerance and quality threshold <> predefined leave triggers for the session (allows you to walk out)
-Walking tolerance (max 10 minutes per hall change).
-Attention span (90 minutes)<> Desired Break frequency.
-My Social energy profile (deep 1:1s vs. big sessions).
-Primary objective and a clear definition of strong ROI for the session

Step 2: Apply Hard Filters: From the Full Agenda Remove Anything That Doesn’t Match Your Interest.

That alone cuts the list dramatically.

No generic “AI transformation.”
No monetization-heavy roundtables.

No sessions without measurable upside.

Step 3: Score a Match Probability and Energy cost For the Rest

For shortlisted sessions, I asked AI to score each session across four dimensions: topic match, depth probability, networking quality, and energy cost.

Energy cost wasn’t abstract. It included hall walking distance, session length, panel format, late-day fatigue, and whether I had a clear leave trigger.

High energy sessions had to justify themselves. Most didn’t.

Step 4: Constrain the Daily Energy Budget

Per day, I allowed at most two medium-energy sessions and one high-energy session (optional). I also required at least one low-energy, high-signal block. Every session has a predefined exit rule. That rule alone prevents sunk-cost bias.

The rest of the time is intentional: expo walks, targeted booth conversations, or breaks.

The result? Across three days at the conference I’m attending only eight sessions. Not because there aren’t more interesting ones, but because only eight meet my defined ROI under energy constraints.

Result: The AI Agent

The interesting part is how you operationalize: How you can turn the logic into a repeatable prompt template — structured intake, scoring model, constraint system, and Day-1 feedback loop. This isn’t about “using ChatGPT.” It’s about building lightweight decision agents without writing code.

For a conference planing workflow, the AI reduces emotional browsing, flags misalignment, and quantifies tradeoffs, while also protecting cognitive energy. The AI does not make the decisions. It enforces constraints.

So when you are building an agentic workflow:

-Define constraints and objectives explicitly and from the get-go. 

-Design and deploy the guardrails before scaling usage, to amplify signal, not noise.

-reduce reactive loops to avoid system and outcome drift.

This constraint-first approach is also how I think about enterprise AI systems — especially where security, reliability, and governance matter.