A tale of two friends exploring AWS AI-DLC
- Raji Krishnamoorthy
- 29 minutes ago
- 12 min read
“Class,” said Mrs. Nair, the Computer Science teacher, “your homework: make a presentation on Software Development Life Cycle with AI. Best one gets a ₹5,000 Amazon voucher.”
Rahul’s eyes lit up. ₹5,000 = cricket bat, shoes, football, racket...
The moment he reached home, Rahul dropped his bag, opened his laptop, and jumped straight into the assignment.
In his mind, the prize money was already half-spent.
• A title slide: “Agile SDLC with AI”
• A neat diagram: Requirements → Design → Build → Test → Deploy
• Some add-on slides about Scrum, sprints, daily stand-ups, retros, etc.
He watched one YouTube video and declared it done.
“Easy,” he said to himself.
Rahul The Overconfident
At 6:30 PM, while the sun was perfect for basket ball, Rahul called his classmate and best friend, Meenu.
“Come to the park,” he said. “Assignment done. Let’s play.”
Meenu was still at her desk, surrounded by notes.
“You finished the SDLC presentation?” she asked surprised.
Rahul raised his eyebrows. “Of course. I am Rahul. Naam tho suna hi hoga..”
Meenu rolled her eyes so hard they nearly rebooted.
In her head: “Congratulations, you are Rahul The Overconfident.”
Aloud, she said, “I’m still working. Teacher said we are in the GenAI era. It is not as simple as you think.”
“Hey, chill Meenu. You know Nair mam has this habit of exaggerating things.” Rahul shrugged. “Just add AI as a helper somewhere near the ‘coding’ and ‘testing’ boxes and it’s done. Okay bye, I’m going to play.”
Meenu angrily looked at her phone and exclaimed “idiot.”
Rahul on the other side played basketball with Rohan and Aditi in the park, came home, had dinner, and fell asleep happily imagining himself walking on to Mrs.Nair, getting the ₹5,000 voucher, and ordering his favourite sports goods from Amazon.
Meanwhile, Meenu’s Strategy
Meenu stared at her blank first slide.
She typed out her title:
“Why software development needs a reset, not another assistant?”
“Hmmm,” she muttered.
“Everyone wants the same things: ship faster, improve productivity, experiment more, release without frying developers’ brains.”
She added bullets:
Faster delivery
More experimentation
Better developer experience
“And now we have Generative AI,” she whispered, typing as she thought. “Tools like Amazon Q Developer and AWS Kiro IDE are already helping with code, tests, docs, migrations… But most teams just slap them into old processes.”
She paused.
“So my story,” she decided, “must explain why that doesn’t work and how AWS AI-DLC fixes it.”
She worked late. Slides grew:
Why retrofitting AI into old SDLC doesn’t work
AI-DLC: AI at the center, humans in control
The new mental model
Three phases: Inception, Construction, Operations
The new language: Bolts & Units of Work
Getting started with AWS AI-DLC
At some point she yawned so wide her brain nearly escaped, set an alarm, and crashed.
Overconfidence vs Anxiety
Morning.
Rahul bounced to the bus stop like a boy who already owned ₹5,000.
Meenu arrived looking like a half-charged phone: functional, but only just.
“Ready to lose?” Rahul grinned.
“You haven’t even seen my slides,” Meenu said.
“I don’t need to. I have Agile SDLC powered by GenAI. Sprints. Stand-ups. Diagrams. Teacher will cry tears of joy.
You probably wrote an essay?”
Meenu clutched her pendrive tightening her jaw muscles and clenching her teeth.
Rahul’s Grand Moment
Finally the moment came for the children to showcase.
In Computer Science period, Mrs. Nair walked in with a serious face and a very dangerous-looking red pen.
“Who would like to present first?”
“Ma’am, I’ll go.”, Rahul’s hand flew up even before the teacher finished her sentence.
He plugged in his pendrive, opened his powerpoint, and turned to the class with a smile that clearly already saw the ₹5,000.
“Good morning,” Rahul said. “My topic is Agile SDLC powered by AI.”
He confidently walked through it:
• “First we gather requirements.”
• “Then we do design.”
• “Then development and testing using AI coding assistants followed by deployment.”
• He drew little arrows and boxes.
• He talked about sprints, user stories, stand-ups, retros.
• He even said “continuous improvement” in a very grown-up tone.
The class clapped. Some kids looked impressed. Rohan whispered, “He is the winner today.”
Rahul straightened up, waiting for the announcement.
But Mrs. Nair’s face was… not “Wow!”It was more “Hmmm… that’s cute but outdated.”
“Thank you, Rahul,” she said. “That is a solid pre-GenAI view of SDLC. Who is next?.”
Rahul froze.
“Pre… what?”
When SDLC Meets AI-DLC
Meenu walked up like a nervous startup about to pitch to a VC.
Slide 1 came up:
“SDLC needs a reset, not another assistant”
She swallowed and began.
“Rahul explained traditional Agile SDLC really well,” she said. “It’s built for a time when:
• Humans do most of the analysis, design, coding, testing, and release work.
• Coordination happens in meetings and other formalities.
• And work moves in long-running cycles — weeks or months from idea to production.
Now everyone wants the same North Star goals or outcomes that Rahul mentioned indirectly: ship faster, increase productivity, experiment more, reduce time-to-market, and not burn out developers.”
She clicked.
“Why retrofitting AI into the old SDLC doesn’t work”
“When we add Generative AI into this old structure,” she continued, “we usually do it like this"
She drew a tiny purple ‘AI’ box squashed somewhere near ‘coding’.
“AI becomes just a fancy autocomplete. So what happens?
AI speeds up producing code and docs…But all that still moves through slow, human-only gates: emails, approvals, meetings.
Teams spend more time re-explaining context to the model instead of capturing it once as shared artifacts.
Or they do ‘vibe coding’: ‘AI, build X’… but there’s no shared spec, no traceability, and big security and quality risks.”
Agile rituals at 2x speed but still the same old dance
All our old methods like SDLC and Scrum were made for slow, long cycles,” Meenu said.
“Work moved in weeks or months, so we added lots of rituals to manage it - daily stand-ups, retros, long planning meetings.
But with AI used properly, work can move in hours or days.
At that speed, you can’t wait for the next stand-up or end-of-sprint retro. You need continuous, real-time checks and feedback. Some of those old rituals start to feel unnecessary.
Story points in an AI world - Grams, Kilos, and a Broken Scale
If AI makes the line between ‘simple’, ‘medium’ and ‘hard’ work blurry, then story points don’t matter as much as before.
And if AI is doing a lot of the boring effort, then maybe velocity is not the main hero metric.
What really matters is the business value we deliver.
AI is also getting better at things we treated as fully manual: planning, breaking work into tasks, requirements analysis, even domain modelling. That means we can go from idea to code in fewer phases.
Not Weak AI, Just an Outdated Track
So the problem isn’t that AI is weak.
The problem is that our old lifecycle was built for a different speed.
With AI, the shape of the work has changed. We can’t just bolt AI onto the side and hope it fits. We have to rethink the process from first principles.
That’s exactly what AWS AI-DLC is trying to do.
AI-DLC: AI at the Center, Humans in Control
She clicked again.
“AI-DLC,” she explained, “is an AI-Driven Development Lifecycle. It:
Puts AI at the center of planning, design, implementation, and operations,
Keeps humans in charge of decisions and judgment,
And uses tools like Amazon Q Developer, Kiro, and agent frameworks as the ‘surface’ where all this happens.
It has two big ideas:
AI-powered execution with human oversight
AI drafts plans, designs, code, tests.
AI is expected to ask clarifying questions and offer options.
Humans approve the important decisions, trade-offs, and risk boundaries.
Dynamic team collaboration
Instead of lone developers struggling silently, teams ‘mob’ around AI output.
Sessions are short, intense decision loops, not long status rituals.”
“So in short,” she smiled,
“AI does the heavy lifting; humans bring context and judgment; the lifecycle is rebuilt so that actually works.”
Rahul shifted in his seat. This was already more advanced than his YouTube video.
The Mental Model: Plan → Probe → Permit → Perform
“Imagine Mrs. Nair decides to build an AI helper for our school plays—a smart assistant on her laptop that can help write scripts, plan roles, and schedule rehearsals. Let’s call it ScriptBot.”
Meenu turned to the board and drew a big loop.
“At the heart of AI-DLC,” she said, “you can picture ScriptBot following one simple pattern again and again. It makes a plan, asks Ma’am questions, she decides, and only then does it start doing the work.”
1. Plan
“First, ScriptBot takes the command from Ma’am, ‘We want to do a play for Annual Day,’ and turns it into a plan,” she continued.
It breaks the outcome into steps:
write the script → choose characters → plan costumes and props → schedule rehearsals → final performance.
It also makes its assumptions explicit: a 20-minute play, one main scene, everyone in class gets at least one line.
2. Probe
“Next, ScriptBot doesn’t just start writing random scenes,” Meenu said. “It asks us questions.”
– Do you want the play serious or funny?
– How many people are comfortable with long dialogues?
– Are there any school rules—time limits or themes—we must follow?
– Do we have a budget for costumes, or should we use what we already have?
“
It gives us Option A versus Option B—like ‘comedy with a moral’ versus ‘full-on serious drama’—instead of silently guessing.”
3. Permit
“Then Ma’am decides,” Meenu said, glancing at Mrs. Nair.
She confirms what ScriptBot got right (“Yes, 20 minutes is fine”) and corrects what’s wrong (“No, we don’t want a sad ending”).
She adds context that ScriptBot can’t know: which student is shy, which jokes won’t work for parents, what the school principal absolutely hates on stage.
4. Perform
“Only after Ma’am is sure about the plan,” Meenu went on, “does ScriptBot get to work.”
It writes the script, suggests who could play which role, drafts a props list, and creates a rehearsal timetable.It keeps everything linked back to the original plan, so if we later change the ending, it updates the scenes and rehearsal plan instead of leaving a mess.
“And this ScriptBot loop,” Meenu said, circling the diagram, “doesn’t happen just once at the beginning. It repeats whenever Mrs. Nair adds a new scene, changes a character, adjusts the length for the time limit, or fixes a problem during rehearsal.”
Rohan leaned over and whispered, “So it’s like a perfect drama partner: ScriptBot does the heavy lifting, and we just keep saying yes, no, and ‘change this bit’.”
Three Phases: Inception, Construction, Operations
Meenu then flashed her next slide:
“The Three Phases of AI-DLC”
With this mental model at its core, “AI-DLC organizes work into three phases,” Meenu said. “They’re not gates but more like layers of shared context.”
1. Inception – From idea to clear units of work
“AI takes high-level goals and drafts requirements, user stories, acceptance criteria, units of work.
The team does Mob Elaboration: everybody reviews AI’s questions and drafts in real time.
Misunderstandings are caught on day one instead of week five.
Output is stored in the repo, not lost in random chats.”
2. Construction – From intent to architecture, code, tests
“AI proposes architecture, domain models, interfaces, and detailed plans.
In Mob Construction, developers and architects steer tech choices, performance constraints, integration patterns.
Tools like AWS Kiro IDE fit here: they start from specs plus Inception context, then generate code, tests, and docs that stay in sync.”
3. Operations – Extend the same loop into runtime
“AI uses the same shared context to suggest infrastructure-as-code, deploy strategies, runbooks, and experiments.
Humans review and approve anything that affects security, reliability, and cost.
The Plan → Probe →Permit →Perform loop applies to deployments, config changes, incident fixes.”
“So across all three phases,” she concluded, “AI doesn’t just answer prompts. It remembers and reuses context stored as real artifacts in the project.”
The new language: Bolts & Units of Work
“In traditional Agile,” Meenu said, “we talk about two-week sprints and huge epics.AI-DLC changes the vocabulary and the shape of the work.”
“First, everything starts with an Intent,” she continued. “That’s the big goal: what we’re trying to achieve in business or technology terms.”
“From each Intent, AI-DLC carves out Units. A Unit is like a small, self-contained slice of the system that actually delivers value on its own. It’s bigger than a single story, but small enough that one team can understand it end-to-end.
Units are designed so they can be built and even deployed independently, without dragging the whole system along.”
“Then, instead of long sprints, we run Bolts,” Meenu said.
“A Bolt is the smallest iteration in AI-DLC.
It’s a short, intense cycle measured in hours or days, not weeks, and it focuses on pushing part of a Unit all the way through build and validation.”
She flashed another slide:
Sprints → Bolts
· Short, high-energy cycles in hours/days.
· Each Bolt has a clear scope and ends with real checks: tests, reviews, quality gates.
Epics → Units of Work (Units)
· Cohesive chunks of functionality derived from an Intent.
· Big enough to matter, small enough to fit into one or a few Bolts, and loosely coupled so they can be built and deployed on their own.
“This isn’t just renaming,” Meenu said.
“It’s a shift from ‘weeks of mixed activity’ to ‘many short, AI-powered Bolts that each move a Unit closer to production, with validation built into every step.’”
Rahul scribbled “Bolts = short build+test for a Unit” in his notebook, still looking slightly worried but a lot clearer.
Where do we start?
Final slide:
“Implementing AI-DLC with Amazon Q Developer & Kiro”
“AI-DLC is just a method unless you operationalize it,” Meenu said.
First and foremost, read the comprehensive AI-DLC white paper that talks about this philosophy in much detail.
Now that you are familiar with the terminologies, it would be an easy and engaging read for you all.
“Two key tools help:
Amazon Q Developer rules – the workflow spine
Kiro – spec-driven, agentic development IDE
Used together,” she finished, “Amazon Q Developer rules and Kiro workflows take AI-DLC off the slide and put it into daily work.”
She looked at Mrs. Nair, then at the class.
“That’s all from me, Thank you.”
Silence. Then — massive applause.
The hollow celebration
Mrs. Nair smiled.
“Rahul,” she said, “your presentation was correct. It showed you understand traditional Agile SDLC well but you were trying to fit in AI.
Meenu, your presentation showed why the traditional SDLC breaks with AI, and how AWS AI-DLC redesigns the whole lifecycle.
So the prize of ₹5,000 goes to…”
Rahul held his breath.
“…Meenu.”
The class cheered.
For a second, warmth rushed through Meenu, her sleepless night and countless powerpoint slides have been worth it. Then she caught Rahul’s expression, his smile slipping before he looked down at his desk.
The 5,000-rupee voucher felt heavier in her hand. She took it with a practiced grin for the photos, but her eyes stayed on Rahul, the celebration suddenly tasting a little hollow.
A Divided Voucher, An Undivided Friendship
At the bus stop that evening, Rahul kicked a stone and said, “Your AI-DLC thing was… actually cool.”
“Thanks,” Meenu replied.
“I thought adding AI to SDLC just meant ‘faster coding’,” he admitted. “Didn’t realize the lifecycle itself had to change.”
Meenu, pulled the voucher out of her bag.
“Rahul,” she said, “this five thousand isn’t just mine. I want to split it.”
He frowned. “Don’t be ridiculous. You won, its your’s”
“I’m serious,” she said. “Either we share it, or it sits in my cupboard and expires.”
He opened his mouth to argue, saw the look on her face, and sighed.
“Fine,” he muttered. “But only because you’re impossible when you decide something.”
Meenu tucked the voucher back.
“Good. So evening we both log into Amazon.in, and order something each. New sketch pens for me,” she replied. “And maybe a notebook and some books.”
“My sports wishlist is already waiting - new football, better shoes, maybe a kit bag”, said Rahul.
And he added, “And we both should practice more of Amazon Q Developer and AWS Kiro IDE. If AI’s doing the boring work, we’ll have more time for games anyway.”
Meenu laughed. “Deal. But no more ‘I am Rahul’ attitude.”
The real lesson
This story is not about topping a class assignment. It is about two aspects:
First, that AI in software is not a cosmetic “helper box” you stick near ‘coding’, but a shift in how work is planned, clarified, confirmed, and implemented end-to-end;
Second, that none of this matters if we forget the humans in the room.
Rahul’s “I am Rahul” swagger and Meenu’s sleepless rigor collide in a way most teams will resonate: old SDLC habits, vibe coding, YouTube-level understanding on one side; deep, slightly anxious rethinking on the other.
The real lesson is simpler.
The teams who will win with AI are the ones who
do the hard thinking on lifecycle, not just tools,
insist on shared context and traceability instead of one-off prompts, and
still choose to split the voucher when the rubric says they don’t have to.
References:
