We are ten minutes in and there is some sort of awkward pause before the session begins amid some conference set-up hustle outside.

The internet access for everyone through the NIMHANS guest profiles has been bit confusing for the most without clear idea. But not so cryptic either to wade through it.

Ashita, the workshop host begins.

Ashita begins with general context and influence on the user experience and defines it with a utility perspectives. But grounds it back into the challenges in building these AI agents.

These challenges include the steep learning curve, the lack of control and flexibility given its non-deterministic nature, the limited visibility and further challenges of building and maintaining.

Ashita introduces strands agents library that allows us to build agents with minimal code.

Further getting into the agentic loop as a definition of the agents that include Model, Tools and Prompts for giving us the results.

Ashita then set the expectations for the days on what kind of exercise the workshop would get us through, the models, evals, guardrails and restaurants and shares AWS's workshop page link

There was a little hiccup with the signing page because people couldn't fully get the query params in the url, but for me, it was the latency in getting OTP and signing attempts failing.

With some help from Ashita, I could get to the exact pages on the AWS Workshop Studio page. The link will stay valid for me for the next 72 hours apparently. Ofcourse, it did not feel like this could be done in couple of hours.

It definitely has been a longer session. Participants were more willing to engage with their anxieties and validations around the future of development.

I stopped taking notes to stay present at the workshop, and I still preferring to do now that workshop is well over 45 min overflown, only spending 45 min on participant engagements.

After spanning and hovering through the using the local model (but in the cloud hosted AWS Jupyter Studio), connecting it to aws services tools, external mcp tools like duckduckgo and finally a guardrails section.

Ashita spent a little more breadth on the gaurdrails section. THe entire jupyter notebooks experience was very straight forward, with a couple bug/flow/changes that need to fixed.

Jupyter made the workshop hence very engaging and slow possibly passive.

Turns out, the session was actually for three hours and till 1pm. I did not specifically gauge the participant crowd. But a bunch of them could be entry level professionals.

Returning back from the light lunch to again remunerate upon the session. And the ideas the host Ashita discussed.

The question hour, which now was almost was for an hour, mainly indulged upon the future of developers. And Ashita had serious applied examples of the change.

The examples that Ashita referred to were Agentic implementations. One of those that she referred to was the personalized UI generation/presentation.

And also, kind of reiterated that, the code writing as a skill, is really now behind. Even the state management. And also the architecture now with spec driven and plan modes.

Or be not the architecture, but the point is traditional software feels like increasingly solved problem. Not entirely sure why agents are then unsolved problems.

Agents then anyway are mostly the unbuilt solutions like much of the software is still yet to be built.

Today's workshop felt like built over the traditional req/res driven software with the streaming abilities for the Agentic integrations.

I will now finally end this session here. The next one is in possibly ten minutes. I will take a break. I might be taking more streamlined notes there unlike the last one.