Director, Software Engineering | Actionable content for leaders in tech. Software engineering, engineering leadership, online business.

15+ years building production systems. Former Director of Engineering at Oracle, leading public cloud services and working with engineers from Microsoft, AWS, Google, Netflix.

My mission: help businesses and engineers build AI systems that create real leverage.

We're entering a world where AI agents do the work and humans direct. The question is which side of the gap you're on.

Tool user: Prompts ChatGPT, starts from scratch every time.

System builder: Has agents that research, write, and code - creating output while they sleep.

Most AI content teaches you to copy paste. I teach the patterns underneath - so you can build any system, for any problem.

One person with the right systems can do the work of 10. I help people become that person.

What I do:
→ AI Engineer: Community for AI engineers. Courses, resources, downloadable AI projects
→ GradientWork: Custom AI development for businesses
→ Newsletter: Free value every week

📩 newsletter.owainlewis.com
👥 skool.com/aiengineer
💼 gradientwork.com

Owain Lewis

5 AI frameworks worth learning in 2026:

(And when to use none of them):

Here's the dirty secret nobody talks about:

AI frameworks are wrappers over provider SDKs with extra features bolted on.

Many professional developers I respect actively avoid them.

The complexity. The over-abstraction. The breaking changes. It adds up fast.

I call this framework tax: the ongoing cost of maintaining someone else's opinions about how AI apps should work.

Start with the SDK. Build until you hit a wall. Then reach for a framework.

When you do, here's what's worth learning:

1. Pydantic AI (Python)

If you've used FastAPI, this feels instantly familiar.

Types, validation, contracts, retries on malformed responses. Works across all major providers.

Best for: Python engineers who already think in terms of types.

2. LangGraph (Python)

Treats AI applications as graphs, not chains.

Branches, retries, parallel execution, human-in-the-loop become first-class citizens instead of afterthoughts.

Best for: Production systems with complex workflow logic.

3. Vercel AI SDK (TypeScript)

Provider-agnostic with a clean developer experience.

Swap models and build agents with minimal code.

Best for: TypeScript developers building AI apps.

4. Google ADK

The first framework that takes polyglot teams seriously.

Official support for Python, Java, TypeScript, and Go with the same mental model. Ships with observability out of the box.

Best for: Teams building on the Google ecosystem.

5. Spring AI (Java)

Exposes LLM services through familiar Spring patterns.

The value isn't innovation. It's integration.

Best for: Enterprise teams already familiar with Spring. 

The uncomfortable truth?

The best framework is often no framework at all.

The goal isn't to master frameworks. It's to build reliable systems.

Sometimes a framework helps. Sometimes it just gets in the way.

What's your current recommendation?

__

Enjoy this? ♻️ Repost it to your network and follow Owain Lewis for more.

Want to master AI engineering?

Join thousands of professionals from top tech companies (Google, Amazon, Oracle) and level up your AI skills for free: https://lnkd.in/e7Ymdh_j

Founder, GradientWork | AI Engineer | I build custom software to make your business more profitable

5 AI frameworks worth learning in 2026

The easiest way to build AI agents in 2026.

(My micro-agent architecture explained):

I built an AI agent that searches YouTube, analyses thousands of videos in seconds, downloads transcripts, writes scripts, and uploads content.

It completely changed my workflow and how I think about "personal AI assistants". I use it every day. 

The whole thing? A folder with some files and about 150 lines of Python code.

Here's the architecture:

1. The folder IS the agent

A Micro Agent is a folder where files define everything:

AGENT md = who it is and what it can do
tools/ = scripts it runs
context/ = knowledge it reads
workspace/ = where it saves files

Point any terminal coding agent at the folder. Claude Code, Codex, Amp, OpenCode. Works perfectly with all of them. 

2. Scripts are a superpower

Software engineers have been writing scripts and CLIs for decades. Nothing new.

But when you give them to an AI agent, magic happens.

Terminal-based agents are exceptional at invoking scripts. This eliminates 90% of the complexity you get with setups like MCP.

No protocols. No complexity. Just stdin/stdout.

3. Skip the infrastructure

The pain of building AI agents is the harness. The UI. The orchestration layer.

There are hundreds of coding agents now. They already have perfect UX out of the box.

Don't rebuild what exists. Focus your energy on what makes your agent unique: powerful tools and rich context.

4. Version control native

Your agent is just files. Share the folder with your team and they can use the agent.

Track changes. Roll back mistakes. Collaborate on improvements. No deployment pipeline needed.

The craft is knowing what you don't need:

Most agents don't need vector databases. Or distributed orchestration. Or complex memory systems.

They need clear instructions, relevant context, and powerful tools that do useful things.

Is this suitable for every case? No. 
Is it suitable for multi user production situations? No
Is it a great way to quickly build useful agents for your own use? Yes.

What's your current setup for running AI agents?

---

Enjoy this? ♻️ Repost it to your network and follow Owain Lewis for more.

Want to master AI engineering?

Join thousands of professionals from top tech companies (Google, Amazon, Oracle) and level up your AI skills for free: https://lnkd.in/e7Ymdh_j

The easiest way to build AI agents in 2026

How I use Claude Code. 

7 tactics to make you a power user:

Claude Code is an AI agent that lives in your terminal. People from all industries are using it for far more than just writing code. Imagine it as the smartest personal assistant you've ever had. 

But the tool alone isn't enough. Here are 9 tactics I use daily:

1. Design before you build (think first)

Separate planning from execution. Use plan mode (Shift+TAB) to think through what you're building before Claude writes any code. What's the approach? What are the edge cases? Think first, code second. This is how senior engineers work. 

2. Break big work into tasks (plan the work)

I take a large spec/plan and automatically break it into smaller tasks to feed to an agent. Each includes what to build, why it matters, and what success looks like. Claude executes one at a time with a clean context window.

3. Iterate, don't one-shot (refine as you go)

No engineer writes code perfectly first time. So why expect AI to? 

After any first draft, I ask: "Review this and suggest improvements" or "What edge cases are we missing?" "How would you improve this code?".

4. Save commands you repeat (systematize patterns)

Use slash commands: I have a .claude/commands/ folder. /design starts a planning session. /review checks code before shipping. /commit writes proper commit messages. Write once, run forever.

5. Set up your project brain (configure once)

CLAUDE {dot} md sits in your project root. Claude reads it at the start of every session. Put your tech stack, code style rules, and things Claude should always (or never) do. You stop repeating yourself. Claude just knows.

6. Automate your standards (automate quality)

I use hooks to run tasks on every edit (tests, linting etc). Quality happens automatically. Problems get caught immediately, not discovered later.

7. Document more things (the mindset shift)

LLMs have no memory. Store important context in static docs. I keep a docs/ folder with SOPs, examples of good output, and rules for how I want things done. 

Claude is a powerful builder with amnesia. My job isn't writing better prompts. It's building docs and systems that make every session start strong.

The future of work = humans using AI agents to become superhuman. 

Which of these are you using yourself?

---

Enjoy this? ♻️ Repost it to your network and follow Owain Lewis for more.

Want to master AI engineering?

Join thousands of professionals from top tech companies (Google, Amazon, Oracle) and level up your AI skills for free: https://lnkd.in/e_KPUM5S

7 Claude Code tactics

Most AI agents never make it to production.

Because no one knows how to deploy them.

Everyone is vibe-coding great looking demos on YouTube.

But how do you take a demo and make it work for real customers?

Most production AI systems fit three patterns:

1. API service

User calls your endpoint, gets a response.

→ Chat APIs for customer support
→ Document analysis endpoints
→ AI-powered search

Use when you need synchronous responses or human-in-the-loop.

Pro tip: Stream the response. Standard request/response feels too slow for LLMs.

2. Scheduled job

Runs on a schedule. No trigger needed.

→ Daily report generation
→ Hourly data sync
→ Weekly newsletter drafts

Use for batch processing at regular intervals. Set it and forget it.

3. Event-driven

External event triggers your automation.

→ New email arrives, process and respond
→ File uploaded, analyse and categorise
→ Webhook fires, take action

Use when reacting to the world in real-time.

The pattern you choose determines your infrastructure.

On Google Cloud, for example:

API service → Cloud Run Service (scales to zero, scales to thousands)
Scheduled job → Cloud Run Job + Scheduler (runs to completion, stops)
Event-driven → Pub/Sub + Cloud Run (decouples source from processing)

AWS and Azure have equivalents. The patterns stay the same.

That's the gap most people miss.

The engineers getting hired aren't the ones with impressive demos.

They're the ones who can answer: "How do we give this to real customers?"

Which pattern are you building with?

__

Enjoy this? ♻️ Repost it to your network and follow Owain Lewis for more.

Want to master AI engineering?

Join thousands of professionals from top tech companies (Google, Amazon, Oracle) and level up your AI skills for free: https://lnkd.in/e7Ymdh_j

3 production AI pattersn

If you want to build AI applications, you need to know how to deploy them.

Most tutorials skip this part entirely.

They end with "and now you have a working chatbot." Running on localhost. Useless for real customers.

Here's what they don't show you:

- TLS certificates
- Load balancers
- Database connections
- Environment variables and secrets
- CI/CD pipelines
- Networking and DNS
- Async work and message queues
- Container orchestration

This is real architecture. Even if a platform team handles it at your company, understanding how software gets to production matters.

I deploy AI applications three different ways depending on the situation:

1. PaaS (Render, Railway)

Push code. They handle infrastructure.

Best for: Side/solo projects, MVPs, "just get it live today."

Trade-off: Costs add up quickly once you scale.  

2. Cloud Platforms (AWS, GCP, Azure)

Managed services with real infrastructure control.

GCP Cloud Run is my default: containerised, scales to zero, pay for what you use. AWS has ECS and Lambda. Azure has Container Apps. All three work - pick one and learn it deeply.

Best for: Client work, production apps, anything enterprise.

Trade-off: Steep learning curve. Each platform has its own approach to IAM, networking, and secrets. Budget real time to learn.

3. VPS (DigitalOcean, Hetzner)

Rent a server. SSH in. Run everything yourself.

Best for: Cost optimisation, GPU workloads, learning how it actually works.

Trade-off: You're the ops team. Security, backups, uptime: all you.

There's no universal right answer.

Shipping a side project? Start with Render.

Building for clients who need scale? Learn a cloud platform.

Want full control and lowest cost? VPS.

Is this the complete picture? No. Every app has different needs.

What's your go-to deployment path?

---

Enjoy this? ♻️ Repost it to your network and follow Owain Lewis for more.

Want in-depth tutorials on this stuff?

Join thousands of professionals from top tech companies (Google, Amazon, Oracle) and level up your AI skills for free: https://lnkd.in/e_KPUM5S

How to deploy AI apps to production

Qoder just launched their Teams Plan which brings agentic coding to enterprise software teams:

One of the challenges with agentic coding is working on larger real-world codebases. 

Qoder is an agentic coding platform that can search and reason across 100,000+ files, giving you deeper codebase understanding - not just individual files.. 

I use it as both a dedicated IDE and a CLI.

Here are the features worth knowing about:

1. Quest mode

This is a killer feature. Describe what you want. Qoder generates a spec, breaks it into tasks, and executes asynchronously.

You delegate, walk away, review completed work later.

This matches how we work as engineers: Plan > [tasks] > execute.

2. Repo wiki

Auto-generates documentation and keeps it in sync as code changes. Architecture overviews, module summaries, dependency maps.

3. Smart model routing

Qoder routes to the right model based on task complexity which makes token usage way more efficient. 

4. Teams Plan

→ $30/seat/month (minimum 2 seats)
→ 2,000 credits per seat
→ Centralized billing and admin console
→ SSO for team-wide access
→ Shared credit packages coming soon

Qoder is moving agentic coding from “nice to have” to serious infrastructure for enterprise software teams.

Check it out here: https://aisecret.co/Owain

If you want to get reliable data from LLMs, learn this essential technique:

The problem is simple:

LLMs return text. Your application needs structured data.

You ask for JSON. Sometimes you get JSON. Sometimes you get JSON wrapped in markdown. Sometimes you get a paragraph explaining the JSON.

Your code expects one format. The model gives you three.

This is why structured outputs exist.

What are structured outputs?

A way to force an LLM to return data in an exact format you define.

Instead of asking the model to return any old answer, you give it a schema. The model's output must match that schema. No exceptions.

When to use them:

→ Extracting information from text (names, dates, entities)
→ Classifying content into categories
→ Routing decisions in AI workflows
→ Any time downstream code expects specific fields

If your AI application needs predictable data from an LLM, don't rely on instructions.

Use structured outputs.

__

Enjoy this? ♻️ Repost it to your network and follow Owain Lewis for more.

Want to master AI engineering?

Join thousands of professionals from top tech companies (Google, Amazon, Oracle) and level up your AI skills for free: https://lnkd.in/e7Ymdh_j

CEO & Founder @CoderFarm | Insurance to Your Project's Success by Providing Specialized Developers & Dedicated Support | React JS & Node JS Resource Augmentation

👋 Hi, I’m Dheeraj — Founder | Builder | Hiring Optimist

The Mission: I am building Coderfarm – A reputation-based hiring platform where startups hire Developers based on Trust & Skills, not just Resumes.

The Backstory (Why I built this): My journey wasn't straight. 🎓 2013: At 19, I lost my father. That grief made me fearless. I promised to build something that mattered. 💼 The Struggle: over the next few years, I worked at 17 different companies. I was searching for a team I aligned with, but I never found it. The hiring system was broken—obsessed with keywords, devoid of human connection.

The Experiment: In 2019, I started my own agency (from a mud-walled, nature-filled office!). We worked with clients like Canva and IIT Roorkee. But I saw the same problem again: Resume Obsession & Misaligned Teams. Hiring is Broken. We are Fixing it. 

The Solution: I stopped complaining and started building. Coderfarm is no longer an idea. It is LIVE. 🟢

We have engineered a Reputation Score system where: ✅ Startups find developers who align with their mission (Culture-fit > Keywords). ✅ Developers prove their skills through real work, not just claims.

What We Offer:

- Verified Trust Scores.

- Purpose-driven matches.

- A bridge between Talent and Opportunity.

The platform is ready. If you are a Founder tired of bad hires, or a Developer tired of being ignored—join us.

Check your Reputation Score here: https://www.coderfarm.in/

Dheeraj Khandare

Is 900 views a failure or a warm-up? 📉

It is Day 6 of building Coderfarm in public. In the last week, we posted 5 videos. The total reach across all platforms? Around 900 views.

For a new creator, that is a lot. For a business, it looks small. But today, Abhilasha asked me the real question: "Views are fine, but do we have traffic on the portal?"

I showed her our Google Analytics. The graph is completely uneven.

And honestly? I am happy about it. Why? Because this is our Baseline. We have done ZERO marketing. No cold calls. No emails. No ads.

We are recording this silence today so that when we switch on the Sales Engine next week, we can measure the real impact.

We are not building for vanity metrics. We are building to prove that Skill > Network.

Take a screenshot of this flat graph. We are just getting started. 🚀
Follow Dheeraj Khandare for the daily climb.

#BuildingInPublic #StartupIndia #FounderLife #GrowthMindset #Coderfarm #Day6 #Entrepreneurship

Founder @ Coderfarm | I help Companies hire Developers who don’t fake their resumes | Building India’s Most Trusted Talent Pool - 100% Pre-Vetted & Verified Talent

The hardest part of building a startup isn't the code—it's the trade-offs.

On Day 7 of building Coderfarm, I hit a wall. I fed my financial data into an AI agent to get clarity, but instead, I got anxiety. Seeing the numbers in black and white made me feel like I was moving incredibly slow.

Here is what derailed my day:
The "Perfect" Job Description: I saw a video explaining that candidates look for future growth, not just a task list. I couldn't stop myself from rewriting our JDs to reflect that, even though I had other priorities.

The Quality Trap: We upgraded our camera gear to improve production value. 

The result? Technical glitches forced us to re-shoot an entire day's video, doubling the workload.

I am constantly fighting the balance between Speed, Quantity, and Quality.
When you are bootstrapping, how do you decide when "good enough" is actually enough?

#BuildingInPublic #Startup #Founders #Hiring #Coderfarm #Productivity

I built a product for users, but forgot the buyers. 📉

Day 4/365 of building Coderfarm.
I looked at my landing page today and realized I made a classic Founder mistake.

I built the page for the people I relate to most: Developers
.
But developers don’t pay me. Companies do. And Universities partner with us.

My landing page was a "mess (a mix) of information. It tried to talk to everyone, and as a result, it spoke to no one.
So today, I tore it down and rebuilt the architecture.

We now have three separate doors:
1️⃣ For Talent: Showcasing skills, not resumes.
2️⃣ For Founders: The hiring solution.
3️⃣ For Institutes: The placement partner.

I also rewrote the HTML semantics from scratch. Because if Google’s bots can’t understand the structure (SEO), a human client definitely won’t find us
.
The lesson? Clarity converts. Confusion kills.

#BuildingInPublic #StartupIndia #UXDesign #SEO #WebDevelopment #FounderJourney #Coderfarm #Indore

The "Experience Paradox" in Tech Hiring: Why are we stuck?

We recently welcomed Rohit Nagar to the Coderfarm team to dive deep into Sales and Marketing. He brought up some critical observations about the current state of IT hiring that every founder and HR leader is grappling with.

Rohit breaks down the major friction points we are seeing today:
1. The Skill vs. Application Gap: Candidates list skills on paper but struggle with real-world execution.
2. The "AI Threat": With AI handling entry-level automation, the bar for "junior" talent has skyrocketed.
3. The Training Cost: Startups are finding it harder to compete with MNCs on salary while absorbing the high cost of training new joiners who might job-hop quickly.

It creates a vicious loop: Graduates can't get jobs without experience, and companies fear hiring without it.

At Coderfarm, we are building solutions to bridge this specific gap—verifying skills so talent meets opportunity faster.

Founders & HR: How are you handling the training vs. hiring dilemma in 2026?

#Hiring #TechIndustry #StartupIndia #HumanResources #Coderfarm #TalentAcquisition #FutureOfWork

Building a team is harder than building a product.

We are a small team of 3—Me, Abhilasha, and now, Rohit. When you are a bootstrapped startup with limited runway, you can't offer big corporate perks. You can only offer one thing:

I asked Rohit to share his honest experience working at Coderfarm. I didn't give him a script. I just asked him what he felt.

His answer hit me hard: "Work feels different when there is a purpose."

We aren't just writing code here. We are trying to fix a broken hiring industry. We are trying to prove that Skill beats Network.

When your team believes in the mission as much as you do, the anxiety of the "unknown" gets a little easier to handle.

Here is Rohit on what it means to solve real problems. 👇

#BuildingInPublic #StartupLife #Coderfarm #TeamCulture #Hiring

When you are a Soldier, you just march. But when you are a General, you have to give the direction. 🧭

It's Day 8 of building Coderfarm. Today, I woke up in panic. I opened LinkedIn and saw I’ve been working on this for 7 years. I opened Instagram and saw 19-year-olds making crores.

Comparison is the thief of joy, and today it stole mine. 📉

I quit my job to find freedom. I thought freedom meant "No Boss". But today I learned the hard truth: Freedom is harder than a job. In a job, someone else takes the blame when things break. In freedom, when the server crashes (like it did today when I moved from Vercel to AWS), or when a user complains about a bug... there is no one else to blame but ME.

I am fixing bugs on code that was working yesterday. I am confused about the direction. But I realized one thing: I asked for this freedom. Now I have to own the responsibility that comes with it.

To all the founders fighting their own mind today: You are the General. The confusion is part of the job description.

#FounderLife #Entrepreneurship #MentalHealth #BuildingInPublic #Startup #Coderfarm #Day8 #Leadership

Hiring Full Stack Developer Intern | Learn AI Coding | Coderfarm

We are looking for a dedicated and passionate Full Stack Developer Intern to join the Coderfarm team immediately. This is a unique opportunity to work on real-world projects and get hands-on training in AI-assisted coding.

🚀 What You Will Learn & Do:
1. Work on Full Stack development projects.
2. Learn how to integrate AI tools into coding workflows (huge career benefit).
3. Collaborate with the core team to build scalable features.

✅ Requirements:
1. Basic knowledge of Full Stack technologies (MERN stack preferred).
2. Willingness to dedicate 8 hours/day.
3. Serious attitude towards learning and delivery.

💼 Internship Details:
1. Duration: 4 Months
2. Stipend: ₹5,000/month (Subject to performance and dedication).
3. Schedule: 6 Days a week | 8 Hours a day.
4. Location: https://lnkd.in/dGpVaADe
5. Perks: Experience Certificate + AI Coding Training.

Note: The stipend and certificate will be awarded strictly based on your dedication, skill application, and successful completion of tasks.

📩 How to Apply: Send your resume and GitHub portfolio to hr@coderfarm.in.

Open article: Coder Farm Pvt Ltd · Indore, Madhya Pradesh, India by share.google

Coder Farm Pvt Ltd · Indore, Madhya Pradesh, India

The hardest part of running a startup isn't writing the code. It's choosing what NOT to do. 📉
It’s Day 5 of my 'Building in Public' journey with Coderfarm.
While the product is stable and the team (Abhilasha, Rohit, and I) is executing fast—fixing bugs on the spot and managing 40K data points—my strategy mind is stuck.
I have 3 paths, but limited bandwidth. I need your help to pick one:
1️⃣ SEO: We need organic growth. But can a bootstrapped team wait 6 months for results?
2️⃣ Viral Tool (Gemini API): Build a free 'Resume Health Check' tool and market it via Reels. High traffic potential, but distracts from the main product.
3️⃣ Direct Outreach: Create a killer 'Demo Video' and send it to Institutes before cold calling. High conversion, lower volume.
If you were in my shoes today, what would you prioritize?
Please drop a 1, 2, or 3 in the comments. I’m reading every suggestion. 👇
#BuildingInPublic #StartupIndia #FounderLife #SaaS #MarketingStrategy #Coderfarm #IndoreStartups

Content Creator & Video Editor Intern | Internship + PPO Opportunity | Coderfarm

Are you a creative storyteller who loves editing? Coderfarm is hiring a Content Creator & Video Editor Intern who can edit high-quality videos and help us grow our brand.

🚀 Key Responsibilities:
1. Edit engaging videos for social media (Instagram/YouTube).
2. Brainstorm content ideas and execute shoots (if applicable).
3. Maintain high quality and storytelling standards.

✅ Requirements:
1. Proficiency in video editing software (Premiere Pro, DaVinci, or CapCut).
2. Creative mindset and attention to detail.
3. Ability to work 8 hours/day with dedication.

💼 Internship Details:
1. Stipend: ₹5,000/month (Performance-based).
2. Future Scope: High chance of a Full-Time Job Offer (PPO) after the internship based on performance.
3. Schedule: 6 Days a week | 8 Hours a day.
4. Location: https://lnkd.in/dagnPZCY

Note: We are looking for serious candidates only. Stipend and Certificate are contingent upon quality work and commitment.

📩 How to Apply: Send your portfolio/best edits to hr@coderfarm.in

taplio

The Battle

Owain Lewis

Dheeraj Khandare

Followers

Posts

Engagements

Likes

Performance ScoreOwain Lewis's LinkedIn Performance

Performance ScoreDheeraj Khandare's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Owain Lewis

Dheeraj Khandare

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Owain Lewis's LinkedIn Performance

Performance ScoreDheeraj Khandare's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreOwain Lewis's LinkedIn Performance