Sign in to view Adeel’s full profile
or
New to LinkedIn? Join now
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
Sign in to view Adeel’s full profile
or
New to LinkedIn? Join now
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
Greater Melbourne Area
Sign in to view Adeel’s full profile
Adeel can introduce you to 10+ people at Commonwealth Bank
or
New to LinkedIn? Join now
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
2K followers
500+ connections
Sign in to view Adeel’s full profile
or
New to LinkedIn? Join now
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
View mutual connections with Adeel
Adeel can introduce you to 10+ people at Commonwealth Bank
or
New to LinkedIn? Join now
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
View mutual connections with Adeel
or
New to LinkedIn? Join now
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
Sign in to view Adeel’s full profile
or
New to LinkedIn? Join now
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
About
Welcome back
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
New to LinkedIn? Join now
Activity
2K followers
-
Adeel Ahmad shared this🖥️ Agents need something that looks like a computer. ▎ 🚫 They don't need an actual computer. ▎ 🧩 My thoughts on capability-based, disaggregated runtimes for AI agents 👇 #AIAgents #SoftwareArchitecture #DistributedSystems #AgenticAI https://lnkd.in/gAJFpuZQThe Agentic Substrate of the Future Is Not a MachineThe Agentic Substrate of the Future Is Not a Machine
-
Adeel Ahmad shared thisAI can write the code. But who decides what to build? My latest article explores what businesses must still own: their recipe and their data—and how I’m building aloop to turn intent into verified code. Models explore. Gates decide. Read: https://lnkd.in/gMCp49h7 #AIAgents #AgenticAI #SoftwareEngineering #OpenSource 🥣AI agents can write the code now. A 1995 Seinfeld episode shows the two things still worth owning.AI agents can write the code now. A 1995 Seinfeld episode shows the two things still worth owning.
-
Adeel Ahmad shared thisAI is everywhere now — in your IDE, in a chat tab, on your phone. Which means this happens more than we admit: You leave work at five, but a hard problem doesn't leave you. That evening you talk it through with an AI on a personal account — no work context, no credentials, just the shape of the problem — and you get somewhere. Decisions. A strategy. Things you've actually settled. Next morning, none of it exists. You re-explain everything to the work tools, re-derive the same decisions, burn the same tokens and hours, and somewhere in the retelling you drop something. The thinking was real; it just couldn't travel. So I built agent-handoff — Git for agents. It packages any conversation, on any platform, into a versioned handoff: what's settled (don't re-establish it), what's open, the tasks in order, with a git-style history the next agent queries instead of browses. One static binary, pure Go, no dependencies. Paste one line into any AI and it packages the conversation you're in. What travels is the reasoning, not the data. The package holds only what you put in it. https://lnkd.in/g2StwVyp Fun fact: this started in the shower. Voice chat with Claude, on my phone, about a problem I'd carried home. We solved it — then built this so the reasoning could reach my work AI in the morning. The first conversation it ever carried was the one that built it. MIT. Feedback welcome. #AIAgents #AgenticAI #OpenSource #ContextEngineering #DeveloperTools
-
Adeel Ahmad shared thisI can look at a container and know what it is before it runs. I can't do that with an AI agent — its model, tools, and network reach are buried in someone's Python. So I built agentrc: a Dockerfile, but for agents. One file declares what the agent is and what it's asking for — a model, a tool, one specific network egress. Every line is a typed request the platform grants, narrows, or rejects, deny-by-default. A security reviewer reads the boundaries before the agent ever runs. It compiles to a normal OCI artifact, so it flows through the pipeline you already have: lint it in CI, pin every skill and MCP server to a digest, review the policy diff like code, ship it unchanged to local, Bedrock, or Kubernetes. Early working draft — tear it apart: https://agentrc.ai What would your pipeline need to check before an agent change could merge? #AIAgents #AgenticAI #PlatformEngineering #CloudNative #AIGovernance #DevSecOps #ModelContextProtocol
-
Adeel Ahmad shared thisSequel to yesterday's post. The Claude Code plugin I shipped — agentic-agile — turns out to be exceptionally efficient. And not just in model spend. In agent productivity: lines of working, tested code per token. One real run. A full Go backend, built from a README and a handful (less than 10) prompts: ~1.5M output tokens 18 packages · ~17K lines · 322 tests · 74 commits · 6 sprints 76% of the work on Sonnet 4.6, 24% on Opus 4.8 99.9% of input served from cache A fully-tested backend for ~1.5M output tokens — most of it on Sonnet, not Opus. Why? Because when you build FOR Agents — instead of retrofitting human tooling onto them — you get exponential results. Agents don't need an IDE. They need: - scoped context — each worker sees its task and its tests, never the whole repo - deterministic gates that verify, instead of asking the model to "double-check" - isolated git worktrees so parallel work can't collide - hooks that make the rules unbreakable, not polite suggestions The model isn't doing the verification — the harness is. So a cheaper model does more, with less context, and physically can't cut corners. That's where the efficiency compounds. Measured on a greenfield build — but the efficiency is structural, so it carries to existing codebases too. Open source: https://lnkd.in/gF2rW6tG #AI #ClaudeCode #Anthropic #AIAgents #TDDGitHub - adeelahmad/agentic-agile: Claude Code plugin for agentic agile + autonomous TDD: hook-enforced AI coding agents that plan sprints, then write code RED→GREEN with deterministic gates. Agentic workflow for LLM developer tools.GitHub - adeelahmad/agentic-agile: Claude Code plugin for agentic agile + autonomous TDD: hook-enforced AI coding agents that plan sprints, then write code RED→GREEN with deterministic gates. Agentic workflow for LLM developer tools.
-
Adeel Ahmad shared thisI shipped my first Claude Code plugin: agentic-agile. The problem it solves: coding agents are eager to declare victory. They skip the failing test, write the implementation first, or quietly "adapt" their way around the process when it's inconvenient. You ask for TDD and get a green checkmark that doesn't mean what you think it means. agentic-agile makes the discipline non-optional. It runs a build as a sprint: a human-gated planning stage, then autonomous execution where every task goes through RED → SCAFFOLD → GREEN, each step in an isolated git worktree, each verified by a deterministic gate wired into the harness — not by the model's own say-so. In Short; Vibe Code with Trust baked in. It's open source. If you're building with agents and want the guardrails to be structural rather than vibes, take a look: https://lnkd.in/gF2rW6tG Install in Claude Code: /plugin marketplace add https://lnkd.in/gF2rW6tG /plugin install agentic-agile@agentic-agile-marketplace #AI #ClaudeCode #Anthropic #AIAgents #TDD #AgileGitHub - adeelahmad/agentic-agile: Claude Code plugin for agentic agile + autonomous TDD: hook-enforced AI coding agents that plan sprints, then write code RED→GREEN with deterministic gates. Agentic workflow for LLM developer tools.GitHub - adeelahmad/agentic-agile: Claude Code plugin for agentic agile + autonomous TDD: hook-enforced AI coding agents that plan sprints, then write code RED→GREEN with deterministic gates. Agentic workflow for LLM developer tools.
-
Adeel Ahmad shared thisStandard fine-tuning is broken in three ways: 1. It memorises — the model copies token sequences instead of learning to reason 2. It diverges — training shifts the output distribution away from the pre-trained baseline 3. It forgets — catastrophic forgetting degrades existing capabilities I've spent the past few months working on an alternative approach: Latent Contextual Reinforcement (LCR). The core idea: instead of rewriting a model's weights, you activate the reasoning pathways it already has. Expert reasoning tokens are interleaved with the model's own generation — the model believes it produced the entire sequence. Gradients flow only through model-generated tokens. The training distribution never shifts because the model learns exclusively from its own outputs. Results on Qwen3-4B, trained on a single machine with 8GB of RAM: → Weight-level change: effectively zero (cosine similarity 1.0000) → Behavioural change: substantial (new reasoning patterns, new identity) → Mechanistic signature: single-axis rotation in activation space → Compute required: a few hours on consumer hardware The mechanistic finding is what I find most interesting: fine-tuning doesn't rewrite neurons. It rotates them. The model is a library — LCR changes which books get read and in what order. This also raises important questions for AI safety. A few-megabyte adapter can modify a model's behaviour while passing every standard evaluation checkpoint. Detection requires geometric analysis tools that aren't yet part of standard auditing pipelines. I've written up the full methodology, experimental evidence, and mechanistic analysis. https://lnkd.in/gCQGySyV #MachineLearning #ArtificialIntelligence #DeepLearning #AISafety #LLM #NLP #ReinforcementLearning #MechanisticInterpretability #FineTuning #MLResearchLatent Contextual Reinforcement: Teaching Language Models to Think Better Without Changing Their…Latent Contextual Reinforcement: Teaching Language Models to Think Better Without Changing Their…
-
Adeel Ahmad shared thisThe Agent Skills spec nails capability definition — but once you have dozens of skills, how do you assemble them into a running agent? I've been building AIOS, an AI-native operating system on Unix primitives, and just proposed extending the spec with AgentFile (declarative agent composition), filesystem-native skill delivery for small models, and multi-agent orchestration via AgentCompose. Would love community input: https://lnkd.in/gpv68rHa #AgentSkills #AIAgents #MCP #ModelContextProtocol #LLM #OpenSource #AIOS #Anthropic #ClaudeCode #DevTools #AIEngineering #BuildInPublic #SmallModels #AgentOrchestrationProposal: AgentFile — Declarative Agent Composition from Skills + Filesystem-Native Skill Delivery [Draft] · Issue #178 · agentskills/agentskillsProposal: AgentFile — Declarative Agent Composition from Skills + Filesystem-Native Skill Delivery [Draft] · Issue #178 · agentskills/agentskills
-
Adeel Ahmad liked thisAdeel Ahmad liked thisYesterdays most amazing launch wasn't Bots in KiroCrew or bulk put items from S3 to DynamoDB... To me it is x86 support in #AWS #Lambda MicroVMs! As of now, you can run x86 MicroVMs in 10 regions: us-east-1, us-west-2, us-east-2, west-1, eu-central-1, eu-north-1, ap-southeast-1, ap-northeast-1, ap-south-1, and ap-south with more regions coming soon. To build your images for x86_64 instead, you will need to add --cpu-configurations architecture=X86_64 #Serverless
-
Adeel Ahmad liked thisAdeel Ahmad liked thisSome chapters are harder to close than others. My eight-year journey at Commonwealth Bank of Australia (CBA) is certainly one of them. I’ve had the privilege of working with some exceptional people, tackling meaningful challenges and building friendships that will last well beyond my time at CBA. The experiences and lessons along the way have shaped me both professionally and personally. A special thank you to Alex, Ranil, Rodrigo and Victoria, and to those who have since moved on from CBA - Andrew, Balaji, Jonathan, David and Dan. Your trust, guidance and support have meant a great deal to me. To everyone I’ve had the pleasure of working alongside, thank you for making these years so memorable. As for what’s next, that deserves a separate post!
-
Adeel Ahmad liked thisAdeel Ahmad liked thisNetflix describes Observability Ontology —an ontology and end-to-end knowledge graph that links telemetry, incidents, deployments, and engineering discussions into a single machine-readable model. The key thesis is that at Netflix's scale, investigations are slowed not by a lack of data or dashboards, but by the absence of formal connections between systems. The main risks are AI hallucinations due to insufficient context, erroneous causal hypotheses, and overreliance on automation. These are mitigated by strict type validation, storing evidence and confidence alongside conclusions, and mandatory human review. https://lnkd.in/eZsmarHM #GenAI #Ontology #KnowledgeGraph #Observability #NetflixOntology-Driven Observability: Building an End-to-End Knowledge Graph at Netflix ScaleOntology-Driven Observability: Building an End-to-End Knowledge Graph at Netflix Scale
-
Adeel Ahmad reacted on thisAdeel Ahmad reacted on thisWe launched a major update to the AWS Fault Isolation Boundaries white paper today. The paper explains the boundaries AWS uses to contain failures: Availability Zones, Regions, partitions, control planes, and data planes. It shows how AWS builds zonal, Regional, and global services on top of them and how we design them to be statically stable during dependency impairments. This design allows you to depend on those services without creating single points of failure in your recovery path. What's new in this version: • Static stability broken down into four example patterns AWS uses in its own services: independent startup, pre-provisioning, maintaining existing state, and asynchronous interaction. • Concrete best practices for zonal and Regional services. These cover observability, recovery tooling, capacity planning, and resilience testing. • Updated best practices for global service dependencies, organized by partitional services, global endpoints, and globally scoped operations. • Step-by-step recovery behavior for Amazon RDS and Aurora, including what does and doesn't depend on Route 53. • Identity guidance for recovery: STS, Organizations, AWS Sign-In, IAM Identity Center, third-party IdPs, and break-glass access. These updates can help you understand where dependencies may hide and how you should expect AWS services to behave during failures. If you run anything on AWS that needs predictable, bounded recovery times, give it a read: https://lnkd.in/e7StbDFcAWS Fault Isolation BoundariesAWS Fault Isolation Boundaries
Experience & Education
-
Commonwealth Bank
***** ******** ******** * ** ************
-
******** ******
**** ****
-
**** *********
***** *** ***** *********
View Adeel’s full experience
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
Welcome back
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
New to LinkedIn? Join now
Licenses & Certifications
-
AWS Certified Solutions Architect - Professional
Amazon Web Services (AWS)
Issued ExpiresCredential ID a71321e9dffb44ce842c8e5f1c70ceb6
View Adeel’s full profile
-
See who you know in common
-
Get introduced
-
Contact Adeel directly
Other similar profiles
-
Vijay Venkatesh
Vijay Venkatesh
I build engines that lead markets and the tech stack beneath them<br> <br>As CTO at Bluesight, I've led a 6× ARR growth journey at a PE-backed healthtech company, executing three M&A integrations, a full cloud migration in 8 months, and a FinOps program that raised enterprise value by $50M. My teams ship AI infrastructure powering the modern hospital pharmacy, faster, leaner, and with a fraction of the turnover. We recently shipped a joint publication with Amazon Web Services on agentic AI in regulated healthcare environments<br> <br>My philosophy: clear vision, clear execution. I thrive in change; scaling a core product, expanding into adjacent markets, integrating acquisitions, and turning technical debt into competitive advantage. I've done it across healthtech, edtech, e-commerce, and logistics including from 4 markets to 40, from 10% conversion to 35%, from 30% attrition to under 5%.<br> <br>What I deliver: M&A technical diligence & integration · Cloud-native architecture · Agentic AI in Production · FinOps & COGS optimization · Full Security program and HIPAA/SOC2 compliance · PE & board-level reporting · High-performing engineering cultures · Machine Learning product delivery<br> <br>"2025 was highly productive, not just from effort, but from sheer results. Vijay has built a prolific engineering organization and a fierce engineering leadership."
5K followersNew York City Metropolitan Area
Explore more posts
-
Dinesh Raghupatruni
Commonwealth Bank • 300 followers
Wrapped up all four Anthropic Claude Certifications #Anthropic #claude If you’re building production AI systems and haven’t looked at these yet, highly recommend testing your skills. If nothing else, it’s a great gut-check on whether you’re using Claude the way it was designed to be used. Claude Certified Architect – Professional https://lnkd.in/gJWqv7-P Claude Certified Architect – Foundations https://lnkd.in/gtsTtV5f Claude Certified Developer – Foundations https://lnkd.in/g-5Tc8Ap Claude Certified Associate – Foundations https://lnkd.in/g-ARs7Rd
43
7 Comments -
Shaju Thomas
Commonwealth Bank • 5K followers
The Evolution of AI Engineering AI engineering is rapidly moving beyond prompts. Prompt Engineering → How we instruct AI Context Engineering → What AI knows Harness Engineering → What AI can do Loop Engineering → How AI works and improves Graph Engineering → How multiple AI agents work together The evolution is clear: From engineering a response → to engineering an agent → to engineering a system of agents. And the next frontier? Engineering the architecture of intelligence. #AIEngineering #AgenticAI #GraphEngineering #AIAgents #GenerativeAI #PromptEngineering #ContextEngineering #HarnessEngineering #LoopEngineering
24
2 Comments -
Muhammad Ali
Langfuse • 7K followers
Excited to be leading a hands-on workshop on Langfuse at AI Engineer Melbourne on June 3-4 at Federation Square. We'll instrument a customer support agent with Langfuse, step by step — tracing, prompt management, online evals (LLM-as-a-judge + user feedback), human annotation, and offline experiments with datasets. The goal is to teach AI engineers best practices for incrementally improving the agentic applications, gaining confidence and increasing shipping velocity. Link in the comments #AIEngineer #Langfuse #LLMOps #AgenticAI #ClickHouse
57
1 Comment -
Pratik Khasnabis
Commonwealth Bank • 518 followers
I work in banking and financial services. There is no secret that CBA is embracing Generative AI in its solutions. However this is from Hacker News today and it is concerning. Mythos hallucinated. 🎯 You can't trust the output from even the most advanced SOTA model and in high stakes workflows a human in the loop review is still essential.
12
-
Anna Leontjeva
Commonwealth Bank • 3K followers
I’m glad to be part of the team developing Brighter AI, CommBank’s new micro-learning series that helps Australians build confidence in using artificial intelligence safely and responsibly. It has been rewarding to help explain what AI looks like in everyday life, from making simple tasks easier to ensuring it is used in a safe, fair and ethical way. Brighter AI began as an internal program to help our people use AI confidently at work. It is now being shared with customers and communities across Australia to support digital inclusion and help more people feel ready for an AI-enabled future. If you are curious about AI and want a simple, practical way to learn the basics, you can explore it here: https://lnkd.in/gvw7ciAU
203
5 Comments -
Matt Cobby
Commonwealth Bank • 5K followers
AI is moving pretty fast — and proximity matters. CBA’s new San Francisco Technology Hub allows us to send teams into one of the world’s most concentrated AI ecosystems where we can access world-class tools, expertise and partners. Not just exploring AI, but embedding teams close to the frontier, learning directly from leading partners, and bringing those insights back to Australia to scale across the organisation. With more 70% of engineering of our 10,000 technologist using AI Power Engineering, CBA is continuing to show what it means to be a leading technology bank — one focused on better engineering, faster learning, and ultimately better outcomes for customers. #CommBankLife #ai #careergrowth #AI #Technology #Banking #Innovation #Engineering #CommBank P.S. I’m impressed by the performance of the latest GPT 5.5 model at so many different situations. Its changing a lot of my workflows and how I approach work.
272
6 Comments
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content