AI-Native Software Engineering
From coding assistants to agentic engineering
Agent loops. Context engineering. Coding agents. Verification-first development.
AI coding has moved beyond autocomplete. Modern coding agents can inspect repos, plan changes, edit files, run tests, and prepare pull requests. The engineers who matter next won't just prompt AI — they'll design agent loops, engineer context, set safe execution harnesses, and verify the work with tests, reviews, and evidence. This is a half-day, hands-on workshop for professional engineers on that discipline — with Claude Code as the working example.
Delivered by
Ritesh Vajariya — on stage, on the road
Two decades in enterprise AI, cloud, and engineering leadership. Keynotes across four continents. Live agentic-AI demos on stage.
AWS Summit
Hong Kong
AI Agents keynote
Live · agentic AI
AWS Public Sector Day
Singapore
Foundation models · framework
Live · enterprise keynote
True Agentic AI · demo
Live · on stage
Enterprise AI keynote
Beijing
AI for industry verticals
Buenos Aires
Collision Conference
Toronto
AI-Native Engineer
Keynote stage
Enterprise AI adoption
Keynote stage
Delivered by
Ritesh Vajariya
Founder, AI Guru · Author, AI-Native Engineer (forthcoming)
Two decades building AI, cloud, and engineering platforms — from Amazon Web Services (AWS) to Cerebras Systems to Bloomberg. Now teaches the discipline behind modern coding agents: context engineering, agent loops, execution harnesses, and verification-first development.
Global keynotes
4 continents
Coursera courses
11+ live
Trained globally
100K+ learners
Why this is different
The engineer's job just changed. This workshop teaches the new one.
Most AI-for-developer content stops at “here's how to ask an assistant to write a function.” That was 2023. Modern coding agents can inspect repositories, propose plans, edit files, run tests, debug failures, and prepare pull requests — end to end.
The engineers who compound value in this era don't “prompt better.” They design the loop the agent runs. They engineer the context the agent sees. They set the harness the agent operates in. And they verify the output before it merges. That is a real engineering discipline, and it's teachable.
This workshop is built for engineers who already ship production code and want to become AI-native — not because AI is replacing them, but because the engineers who own the loop, the context, the harness, and the verification will out-ship everyone else.
The future engineer is not replaced by AI. The future engineer manages AI work.
— Ritesh Vajariya, AI-Native Engineer
The signature framework
The AI-Native Engineering Stack
Four pillars. Teachable in one glance. Durable for the next decade of software engineering.
Context
What the agent knows
The bottleneck isn't better prompts — it's better context. What's the repo shape? What are the coding standards? What are the failing tests telling us? What's the architecture doc say?
- →Repo maps, ADRs, and architecture docs
- →Coding standards + prior decisions
- →Failing tests, logs, and traces
- →AI-ready engineering briefs
Loop
How the agent works
Read → plan → act → observe → verify → revise. A disciplined loop turns coding agents from party tricks into production tools. This is where the engineer stops being a coder and starts being a supervisor.
- →Plan-first workflows before touching code
- →Small diffs, tested incrementally
- →Explicit acceptance criteria per step
- →Stop conditions and evidence bundles
Harness
What the agent can do
Boundaries beat instructions. File permissions, approval gates, cost + retry limits, network policy — the harness is what keeps agent autonomy safe.
- →File and shell permissions
- →Approval gates for risky actions
- →Cost, retry, and time budgets
- →Sandboxes and network policies
Verification
How we prove it worked
Tests as agent contracts. Reviews as evidence. In AI-native engineering, tests aren't just quality control — they're how humans communicate constraints to agents.
- →Tests as first-class specifications
- →Static, dynamic, and security checks
- →PR-ready evidence bundles
- →Human review of the diff, not the story
Weak agent instruction
Build login feature.
One line. No context. No stopping condition. Coin-flip results.
Strong agent loop
wait for approval → implement small diff →
run tests → fix failures → summarize
evidence → stop when acceptance criteria pass.
Context, loop, harness, and verification — all four, all present.
Half-day agenda
Four hours · Framework · Hands-on · Career playbook
Roughly 50% lecture (framework + worked examples) and 50% hands-on (Claude Code agent loop, context engineering exercise, PR evidence bundle).
| 0:00–0:20 | The AI-Native Engineer — From Coder to Agent Supervisor |
| 0:20–0:55 | The New Stack — Context, Loop, Harness, Verification |
| 0:55–1:45 | Hands-on — Build a Feature with Claude Code Using an Agent Loop |
| 1:45–2:00 | Break |
| 2:00–2:35 | Context Engineering — Repo Maps, ADRs, and AI-Ready Briefs |
| 2:35–3:10 | Verification-First Development — Tests, Reviews, Evidence Bundles |
| 3:10–3:40 | Agentic CI/CD — Where AI Goes Beyond Writing Code |
| 3:40–4:00 | Career Playbook — Becoming an AI-Native Engineer |
The hands-on centrepiece
You'll build a real feature with a disciplined agent loop
Not “ask Claude to build X.” A supervised, tested, PR-ready feature — with evidence at every step.
The disciplined Claude Code workflow
- 1Start with a feature request
- 2Ask Claude Code to inspect the repo
- 3Ask it to summarize the architecture
- 4Ask it to propose an implementation plan
- 5Approve only a small, well-scoped change
- 6Let it edit files
- 7Run tests
- 8Inspect any failure
- 9Ask it to fix
- 10Ask for a PR summary and evidence bundle
- 11Review the diff manually
Claude Code is powerful. The engineer owns the loop.
That is the message. That is the discipline.
Live agentic-AI workflow · LiteLLM traces on stage
Recent keynote · 2025
Proof of practice
Not theory — Ritesh runs the same discipline live on stage
Frame from a recent keynote: a full agentic AI workflow running with real LiteLLM completion traces projected on the screen behind him — the same verification-first pattern this workshop teaches, in front of a live audience.
What you'll learn to do in 4 hours is what Ritesh already does on international keynote stages.
Tools you'll work with
What you'll walk away with
Six outcomes for engineers who ship real code
Not abstract. Not aspirational. Concrete patterns you can drop into your team's workflow the next day.
Design an agent loop from scratch
You'll leave able to structure a coding-agent task as a disciplined loop — plan, act, verify — not a one-shot prompt. This is the single biggest lever most engineers are missing.
Engineer context that actually helps
Turn a bad task ('add payment support') into an AI-ready brief with the right repo maps, ADRs, tests, and constraints. This is where 10× productivity gains come from — not clever prompting.
Set safe execution harnesses
File permissions, approval gates, cost limits, and sandboxes. When agents run without adult supervision, harnesses are what keep production safe.
Verify AI-written code you didn't write
Tests as agent contracts. Evidence bundles at PR time. Review the diff, not the story. You'll leave with patterns you can drop into your team's workflow the next day.
Work Claude Code like a senior engineer
Not just 'ask Claude to build X.' Real workflows: inspect repo → propose plan → approve small diffs → run tests → fix failures → PR summary. Hands-on for most of the session.
See where agentic CI/CD is going next
Beyond writing code: dependency upgrades, incident triage, log analysis, release notes, migration support. Where AI goes when the loop leaves your laptop.
Tangible takeaways
What you keep after the session
Every attendee receives
AI Guru® Certificate of Participation
Signed by Ritesh Vajariya, Founder of AI Guru. Issued digitally after the workshop — LinkedIn-shareable, printable, verifiable, and yours to keep.
Preview shown — your certificate will carry your name and the event date.
Curated audience
Who this session is for
A room of engineers who already ship real code — so the hands-on time actually lands.
Non-technical professional interested in AI? See our AI for Working Professionals track →
Questions
Frequently asked
Do I need to know Claude Code, Cursor, or Copilot beforehand?+
No prior experience with coding agents is required — we set up the tools together. What we do assume: you already write software professionally, use git, understand tests, and have opinions about code quality. This is not a first-AI session; it's the discipline layer on top of one.
What programming languages and stacks are covered?+
The framework — context, loop, harness, verification — is stack-agnostic. Hands-on demos use a mix of Python, TypeScript/Node, and Go; the patterns apply equally to Java, Rust, C#, etc. Bring your own laptop and preferred language.
Is this hype about 'AI replacing engineers'?+
The opposite. This session is grounded in the position that AI does not replace engineers — it changes what engineers do. The future engineer is not a prompt writer; the future engineer is an agent supervisor, context engineer, and owner of correctness. That is the message, and the workshop teaches the discipline behind it.
How much of the session is hands-on vs lecture?+
Roughly 50/50. The framework sessions (Context, Loop, Harness, Verification) are taught with worked examples and live diffs. The 50-minute Claude Code block is fully hands-on — you build a real feature with an agent loop and take away the working project.
What should I bring?+
A laptop with git, Node.js 18+, and a code editor (VS Code, Cursor, or JetBrains). We'll help you set up Claude Code, Cursor, or a Copilot equivalent during the first hands-on block. Bring a repo you know well — you'll get more out of the exercises applying them to real code.
Is this only for senior engineers?+
No — but you should already be comfortable writing, testing, and shipping software. The framework becomes more valuable the more code you own, but the discipline is teachable at any level. Advanced CS / IT students who already ship real projects will get significant value.
Will there be a recording?+
No live recording distributed publicly. Attendees receive slides, worked examples, the reading list, and the working Claude Code project. Video segments may be re-used in later AI Guru content — attendees will be asked before any personal footage is used.
Is there a fee?+
The Ahmedabad pilot pricing is being finalised — a small nominal fee to keep the room committed, or free for a limited number of invited engineers under a partner arrangement. Register your interest and we'll share the pricing before it's public.
Will this be run in other cities?+
Ahmedabad is the pilot. Based on response, we plan to extend this specific developer-focused workshop to Bangalore, Pune, Hyderabad, and NCR through 2026. The broader multi-track workshop series covering both engineers and working professionals continues across seven tier-2 Indian cities.
Can my company host this internally for our engineering team?+
Yes. If you'd like to run this as an internal engineering-team session at your company (30–100 engineers), we can customise the agenda around your stack, your codebase, and your current AI tooling. Reach out via the partner CTA on this page.
Ahmedabad pilot · 2026
Reserve your seat in the pilot cohort
The Ahmedabad pilot runs first — 40–60 engineers, half-day format, hands-on with Claude Code. Register your interest and we'll send date, venue, and pricing before it's public.
Or email us directly at [email protected]