Research.

Who Trains the Next Senior?

Managing Software Engineering When AI Makes Code Cheap

Sairam Krishnan, Gajendra Babu Thokala

Generative and agentic AI have made the routine work of software engineering nearly free, and the first casualty is the junior engineer. Cheap machine-made code commoditizes breadth while raising the value of depth, the judgment of which answer is right in this system, and divergence, the perspective from outside the consensus that a senior-plus-AI team structurally lacks. Left to inertia, organizations replace juniors and let review and apprenticeship hollow out. The paper lays out the alternative: shrink the unit of review, tier scrutiny by consequence, measure verified outcomes rather than raw output, and rebuild the apprenticeship on purpose.

A Multi-Layer Framework for Evaluating the Return on Investment of AI Projects

Sairam Krishnan, Andrea Hepp, Sonia Gandhi

More than 80 percent of AI projects fail to deliver measurable returns, and traditional capital budgeting misses what sets AI apart: deeply uncertain costs, indirect and non-linear benefits, path-dependent value creation, and acute sensitivity to organizational readiness. The framework integrates four layers into one investment decision process: a SCALE strategic readiness assessment, ROIC decomposition through a value driver tree, stage-gate phased investment, and a synthesized decision scorecard. An eight-category risk-adjustment table sets the discount rate for AI-specific risks, including vendor and platform dependency and organizational friction. The framework is validated against eight deployments: Salesforce Agentforce, Palantir, Spotify, Duolingo, Shopify Magic, Uber, JPMorgan COiN, and NatWest.

Drone-Megaddon 2026

Co-Intelligent Swarm Control with Language and Vision Models

Sairam Krishnan, Daniel Ting

Twelve years after the original interface, commercial drones ship with LTE and GPS-denied navigation, drone warfare has produced public lessons on electronic warfare and operator load, and foundation models have made natural-language intent and visual grounding workable interfaces to robotic systems. The paper keeps the 2014 core of RTS selection, the $RAD command queue, ACK’d radio, and human oversight, and proposes three AI layers over it: an intent translator that turns an operator’s words into a validated command sequence, a perception grounder that binds a phrase like “follow the red sedan” to a tracked target in a drone’s video feed, and a coordination planner that gives every drone its own final position. It walks through a search-and-rescue example and addresses the ethical questions the 2014 paper did not.

Drone-Megaddon

A Real-Time Strategy Interface for Managing Drone Swarms

Sairam Krishnan, Daniel Ting

Researchers had controlled single drones through neural sensors, Oculus Rift, and mobile devices, but no adequate interface existed for commanding many drones at once. Drone-Megaddon is an Android real-time strategy interface patterned after StarCraft II, built end to end as a cyberphysical system: the operator selects drones, issues a high-level command, and the swarm executes it.