lukla.logic Engineering partner
Start a project →
LUKLA_LOGIC/ INSIGHTS
Insights · Field notes from the practice

What we're
thinking about
this quarter.

Field notes from the practice - what we're learning while shipping production systems with agent fleets. No SEO trend pieces. No top-10 listicles. Just things worth writing down.

SR.ENG AGENT.ARCHITECT AGENT.FRONTEND AGENT.BACKEND AGENT.TESTS AGENT.RELEASE FIG · ORCHESTRATOR + FLEET
★ APPLIED AI 8 min read May 27, 2026

Why AI Projects Fail After the Demo

Most AI prototypes look impressive in week one. The hard part is turning the demo into a reliable system that survives real users, messy data, and production constraints.

SN Lukla Research · Sudarshan Neupane, Founder
Read the essay →
02 / WRITING

Recent.

11 posts · sorted by recency
FEATURED · ESSAY18 minMay 18

A new era of engineering delivery.

The conventional engineering team is not the natural shape of software work. It is a coping mechanism for the era when one person could not hold the whole system.
Bibek Pandeyread →
AGENTIC9 minMay 12

One mind, a fleet underneath: orchestration patterns for senior engineers

Six patterns we use for splitting work between an engineer and a multi-agent fleet, and the patterns that look right but quietly fail.
Bibek Pandeyread →
APPLIED AI14 minMay 04

RAG is not a feature. It is an engineering discipline.

Most retrieval-augmented systems fail not at the model layer but at the engineering layer around it. A field-tested checklist.
Anjali Maharjanread →
OPERATIONS7 minApr 29

The 24-hour change cycle: how we close most requests by next morning

It is not magic and it is not working harder. It is a specific operating model.
Prabin Shrestharead →
ENTERPRISE11 minApr 22

Strangling a 14-year-old monolith, politely

How we moved a regulated insurance platform off its mainframe without one big-bang weekend.
Tara Adhikariread →
AGENTIC5 minApr 14

What we do not ask agents to do

A short list of the architectural and product decisions we still will not delegate.
Bibek Pandeyread →
PRACTICE6 minApr 06

Outcome pricing: numbers that ship

Why we do not sell hours, how we scope outcome contracts, and what we tell clients who insist on per-hour pricing anyway.
Founder's Officeread →
APPLIED AI10 minMar 28

Evals as tests: bringing CI/CD discipline to AI features

If you would not ship code without tests, you should not ship a model without evals.
Anjali Maharjanread →
OPERATIONS4 minMar 19

Quiet alerts: noise-free incident response

A page that pages too often pages no one. Our rules for what counts as an alert and what does not.
Prabin Shrestharead →
PRACTICE12 minMar 10

Pushing back: the part of the job we charge nothing for

Why we tell clients when the feature is wrong, before we write a line of code.
Founder's Officeread →
Field notes

One letter a month, from the practice.

Long-form essays on agentic engineering, applied AI, and the operating model of modern software. No marketing. No more than once a month.