Speak at The Fifth Elephant 2026 Annual Conference
Share you work with the community
Jul 2026
13 Mon
14 Tue
15 Wed
16 Thu
17 Fri 09:00 AM – 06:00 PM IST
18 Sat 09:00 AM – 06:00 PM IST
19 Sun
Tanvi Bhakta
@tanvibhakta
Submitted Jun 21, 2026
Picture this: you’re building a complex pipeline of business logic. Of course, it is 2026, so you’re not writing code by hand.
You build the [piece A] of it, and then later build [piece B that interacts with A] - except the LLM has gone and [interacted with piece A in a different manner to how you would have done it but achieving the same result]. Existing verification methods validate
But the control flow of the program has drifted from what the author intended.
Most existing frameworks coalesce around allowing humans to do many kinds of things on top of the framework; but agents, across multiple sessions, cannot hold [the memory of] a single mental model in its context. It reads the files necessary for the work to be done and makes architectural decisions that fit the local maxima of the context it has. Unfortunately, this often results in creating hard-to-reason about spaghetti code that does not match the author’s intention.
For agentic development of software to move from vibe-coding back to software engineering, it is essential that we architect our software to have very strong opinions. There should only be one way to do one thing, and it should be easy for the human to skim and develop confidence in what is happening.
The rest of my talk will focus on HOW we can make this possible - DAGs to encode business logic, thick abstractions over many layers, state machines to drive UI, etc.
My session is going to be beneficial to
I’m a founding engineer at alt.inc
TODO:
{Add the link to draft slides - PDF/PPT - with comments access}
{Add the link to 2-min elevator pitch video}
{{ gettext('Login to leave a comment') }}
{{ gettext('Post a comment…') }}{{ errorMsg }}
{{ gettext('No comments posted yet') }}