After one session, one member said: "This was more valuable than a $1k+ Maven workshop."
Each month, updates in the AI space leave everyone asking:
Is my job at risk? Or is this creating more work slop for humans to clean up?
Will the latest machine learning models cure cancer or just generate TikTok slop?
How far am I behind if I don't know the acronym soup that is MCP, A2A, GPU, TPU, NPU, VLA, VLM, LLM, RLHF, ROFL etc?
The answer is somewhere between these extremes.
It's important to stop scrolling headlines and start building with the latest tools to seek understanding. With hands on experience, it's easier to get a sense what's possible. Experimenting also helps create a clearer path to see where the future is headed.
It's never been easier to start to build, 25minutes gets a production level agent workflow for simple tasks on platforms like Lindy.
The problem is most people don't know where or how to start.
We hosted a 3 part build along to solve this problem
"The structure was great - walking through the steps and explaining your reasoning at each stage."
In one month, we had 30 MIT Sloan alumni join for a three part build series. Many had never built an agent before. No one needed 20hrs of lectures for RLHF, debating model benchmarks, or to wait for 'AGI.'
To cut through the noise of marketing, we helped everyone focus on one thing: build an agent for a usable output. We built with Lindy because it's the easiest way to build AI agents and workflows.
Everyone shipped agent workflows that they can use on different tasks:
Calendar assistant for meeting prep and automatic updates
Research agent for contact enrichment and update CRM data
PM agent to orchestrate subagents for collecting vendor information
Competitive intelligence through scraping socials and websites
Creating dashboards through analyzing data with Computer Use
The main principle here is build first, learn third.
It's counter to conventional wisdom in education which values theory over experience making learning the objective function. It would be like learning how to walk by going through 14 weeks of lectures and doing problem sets on the mechanics of walking.
Walking is best figured out through trial and error. Fall down. Get back up. With tools like Lindy or Claude Code, everyone has the ability to walk right into their first agent. Exploring what's possible by building is where the real learning happens.
What we're seeing:
Build first. This can be 1-3hr over a coffee, or a 10hr weekly sprint to figure out what's possible. For example, Shopify has been leading the way for their team with internal experiments shared at town halls.
Next, figure it out. Agents are finicky right now. It's more an art than a science, like managing humans because the foundation models are natural language, large language models. This takes time to iterate and test what works. The meta-learning is you can use agents to build agents. Try out Lindy AI Ask. Note, they aren't great yet, but they help get over the blank canvas problem.
Learning happens third. The best learning comes from this experience. Reflecting on what works and what doesn't. Thinking about next steps after units of build time.
Here's how we enabled that build first culture.
First, session design
Lindy template for Build #2 / Customer Support Agent
"Thank you so much! This was so helpful."
The core of these workshops are the agents that get built. One layer around the build is a fundamental skill. Around that is a case study which helps make it real.
We combined
Case study - understand real world application for how others are using tools
Fundamental - understand core concepts for what and why agent workflows work
Build Along - guided warm up and sprint to ship by the end of the session
For session #2 what that looked like:
Case study: Covering Klarna's 1 year experiment for going in on AI agents for customer support. Spoiler: human in the loop is best practice.
Fundamentals: Understanding how manual chain of thought prompting can context future steps. Levels of intelligence, where even self driving cars are not level 5 autonomy right now. Human + AI is a great combination.
Build along: Screen shared above. We built a sandbox agent, using web chat, and simulated dataset to test out a customer support agent. The main
Building the agents were super fun. Each agent template was stress tested and evaluated to make sure it was robust enough to solve for different use cases within the scenario.
The virtual session experience
Each session included a follow along guide and template agent to give a strong starting point. The session itself was dynamic as we adapted to questions as they came up. We spent 90% on the Lindy editor canvas to keep it focused towards building.
Agenda
Intros
Case study
Fundamentals
Agent warm up
Build along sprint
Demos + Debrief
Each one is packed as dense as possible with insights to make the experience valuable and recording worth going back to.
Highlight from the series: one tool I find most interesting right now: computer use
Computer use spinning up code in terminal within Lindy task manager
"I really learned a lot."
In the third session, we made it to more of an open build than a follow along. We first introduced how the computer use agent works by spinning up a virtual machine. Most people think browser use, like "oh, this uses chrome." But on Lindy, computer use means access to terminal, file system, and browser. This means it can handle more complex tasks.
For the demo, we took in workshop data and let the agent figure out the best pre-session dashboard. One interesting feature of this agent is the self testing feature. Once it designs something, whether it's an error in code or in clicking through browser, it figures out how to fix it.
Check out Lindy docs for more info and ideas.
With the bar for what's possible changing every 30-90 days, building is really the only way to understand what can be applied right now.
Tl'dr
In the age of accelerated progress with AI, it's important to stop scrolling headlines and start building with the latest tools to seek understanding.
Build first, learn third. Go through the experience of making something, figure it out, and reflect after.
Case studies and fundamentals help set a baseline for what's being built. Just in time information with live QA helps fill in the gaps to help people level up
It's never been easier to build with new tools like Lindy or Claude code
Try out Lindy Computer Use. It's really interesting as a virtual machine with terminal access, file system, and browser use. You can spin up in 1 click and all programmed in natural language.
What's next for us
This build along series went well and was fun!
We're looking to offer this for more communities. If you have a community you're a part of, like a professional group, alumni base, or internal team, reach out.
We're starting to design role specific workshops. This was an intro, and now we're exploring how to bridge the gap to implementation. For example, what would an AI agents for PMs build along look like as a part 2 to this series? How do agents enable and empower RevOps teams? What role will building MCPs play in future of self driving labs?
Shout-out to Nabil for the collab on this series!