About
An accomplished product strategy and technology leader, I have 20+ years’ experience…
Articles by Paul
-
Forget about being CEO of the product, Product Managers should act like Founders.
Forget about being CEO of the product, Product Managers should act like Founders.
Several weeks ago I was invited to speak at a Product School event on the topic of Conscious Product Management. The…
Activity
2K followers
-
Paul Greenwell posted thisSix months into your AI transformation, can you actually show what changed? Most enterprises can't. Real money goes into tools, training and new ways of working — and the executives funding it are steering on stories, not evidence. Our Lead Architect Abdi Daud has written up how we measure the shift, and it cuts against most of the advice out there: → Raw output metrics (lines of AI code, acceptance rates) answer the wrong question. Measure the shift toward agentic development, not tool usage. → Don't impose a new metrics regime. Report against the delivery metrics the org already trusts — and retrofit the baseline from history that already exists. → AI-assisted delivery compounds. Judge it at week four against a promise of instant transformation and you'll kill programs that were about to pay off. → The telemetry is nearly free. The gap isn't data — it's connecting that exhaust to a narrative executives can steer by. If you can't show the baseline, the shift, and the trend, your AI transformation is a leap of faith wearing a business case. Worth 5 minutes: https://lnkd.in/gnnMEarz #AITransformation #EngineeringLeadership #AIEnablement
-
Paul Greenwell shared thisAI has made building software dramatically cheaper. That's exciting. But cheaper to build doesn't mean better to use. When implementation was expensive, it forced discipline. Teams debated trade-offs, killed mediocre ideas early, and said no to features that weren't worth maintaining. That friction was a feature, not a bug. Now that friction is disappearing. And without product discipline to replace it, we're heading toward an era of "AI Slop" — bloated, generic software that technically works but nobody actually wanted. The 2025 DORA Report nailed it: AI doesn't fix a team. It amplifies what's already there. I wrote about it: "Coding Is Getting Cheaper. Thinking Never Will." https://lnkd.in/gV3nUUiW #ProductManagement #AI #SoftwareEngineering #ProductThinking
-
Paul Greenwell shared thisVibe coding has its place. It's great for prototyping, hacking together a proof of concept, exploring an idea quickly. But for the companies we work with — product companies scaling fast, enterprises managing complexity — it was never going to be enough. When you're building software that needs to scale, perform under load, meet compliance requirements, and pass security audits, you can't just "vibe" your way through it. I've watched product teams and enterprise engineering orgs adopt AI-assisted development over the past year. The pattern is consistent: the ones that treat AI like a shortcut hit a wall. The ones that treat it like a new delivery capability keep accelerating. The difference is where AI shows up. Not just in the code editor, but across the entire product delivery lifecycle — from requirements analysis and architecture decisions through to test generation, code review, and release validation. We're calling it an AI-augmented PDLC, and it changes the economics of how product teams operate. Vibe coding optimises for speed of first draft. An AI-augmented product model optimises for speed of shipping something you'd actually stand behind in production. For product companies scaling fast, this is a competitive advantage. For enterprises managing complexity, it's becoming a necessity. Vibe coding was the prototype. An AI-augmented PDLC is the product.
-
Paul Greenwell shared thisWhat's the actual ROI of AI? It's the question every executive asks — and the answer used to be vague. Not anymore. Here's what we've delivered for Australian businesses in the last 12 months: - Agentic processing system: 2x output, 32% cost reduction - AI-powered design workflow (fashion): 60% time saved in sampling, 2x creative output - AI video classifier (advertising): 75% time saved, 3x output - Agentic financial analyst: research turnaround from weeks to minutes - AI voice agent (call centre): 4x productivity, 90% cost efficiency improvement - AI pricing agent (hospitality): time-to-recommendation from hours to 3-5 minutes None of these were moonshots. Each was delivered in 6–12 weeks for $40–100K. The pattern? Start with a specific workflow. Measure before and after. Ship fast. Iterate. The companies seeing results aren't the ones with the biggest AI budgets — they're the ones picking the right problem first. What workflow would you start with? Drop it in the comments — happy to share what we've learned.
-
Paul Greenwell shared thisGot a notice from Azure yesterday: GPT-4o realtime and audio preview models are being retired on 24 March 2026. Two weeks' notice to migrate production workloads. This is the new reality for any team building on AI. Models aren't static infrastructure — they're living dependencies with their own lifecycle, and the deprecation cadence is accelerating. Most organisations we work with at Propel are still treating AI models like they treated databases five years ago: set and forget. But the AI stack moves faster, and if you don't have a model governance strategy, you're accumulating a new kind of tech debt. A few things worth having in place: → An abstraction layer between your application logic and the model API → A testing framework that can validate behaviour across model versions → A clear owner for AI model lifecycle within your engineering org The organisations that will win with AI aren't just the ones that adopt it fastest — they're the ones that can manage the change it brings. #AI #TechDebt #EngineeringOps
-
Paul Greenwell shared thisThe biggest AI risk in most organisations isn't the technology. It's the leadership gap. We've been running AI masterclasses for executive teams, and the pattern is consistent: most leaders know AI is important, but very few have a practical mental model for how it changes their business decisions. That gap creates two problems. First, it slows down adoption because every AI initiative needs to be "sold" internally. Second, it leads to poor investment decisions — either over-investing in hype or under-investing in genuinely transformative use cases. The fix isn't another strategy deck. It's hands-on exposure. Leaders need to use the tools, see what's possible, and build their own intuition for where AI creates value in their specific context. If your leadership team hasn't had a structured AI immersion yet, 2026 is the year to do it. The gap between AI-literate and AI-illiterate organisations is widening fast.
-
Paul Greenwell posted thisLast night I watched the memorial for the Bondi attack. It was dignified and deeply moving, particularly when Rabbi Yehoram Ulman, the head of Chabad Bondi, read the names of each of the fifteen victims and spoke about what we can learn from each of them. Earlier in the service, the Prime Minister was booed — a moment that reflected the depth of anger and fear many in the Jewish community are carrying right now. While I don’t agree with it, I understand why it happened. That reaction reflects a growing frustration over what many see as a lack of leadership since October 7. After the Hamas atrocities in Israel, protests began in Sydney and Melbourne before Israel had even entered the war. Since then, our CBDs have regularly been shut down by demonstrations that are not just anti-Israel, but often openly anti-Australian. Chants like “from the river to the sea” and “globalise the intifada” have been tolerated. We have now seen in Bondi what that rhetoric can translate into. During this period, the Foreign Minister was criticised for not visiting the Israeli communities devastated on October 7, instead choosing to visit the PA — a decision many in the Jewish community experienced as a lack of solidarity. The government has since recognised a Palestinian state despite the absence of a credible partner for peace. There is also an uncomfortable truth that often goes unacknowledged: anti-Zionism is experienced as antisemitism. Zionism is simply the belief that Jews, like any other people, have a right to self-determination. While some argue these ideas can be separated, in practice they are not. Jews are not being targeted over nuanced geopolitical views, but because they are Jewish — as seen in the attacks on Jewish businesses, synagogues and daycares here in Australia. For me, this isn’t abstract. My kids attended a Jewish school their entire schooling. Armed security was always present and unfortunately was just normal for them. Hanukkah in the Park at Caulfield Park used to be something we looked forward to every year. It doesn’t happen like that anymore. Jewish events are increasingly held behind security, with locations shared only at the last minute. This has become routine. The Bondi memorial was uplifting in its resilience. But it also reinforced a hard truth: the foreseeable future looks like more security, more restrictions, and less openness. That should concern every Australian — not just the Jewish community.
-
Paul Greenwell reposted thisPaul Greenwell reposted thisWe’re on the hunt for a senior Customer Success leader who gets enterprise clients, loves solving real problems and can turn the hard work done by our delivery teams into strategic partnerships. If you know someone who’d thrive in our product-led consultancy, we would be grateful of you could send them our way.
-
Paul Greenwell shared thisWe’re hiring! Propel Ventures is looking for a Senior Software Engineer to help build innovative products that drive real impact — if you love solving complex problems and working in a collaborative product-led environment, we’d love to hear from you.
-
Paul Greenwell liked thisPaul Greenwell liked thisWe thought we’d get into the Grand Final spirit at NWS… 🔴🔵The boss hasn’t approved the new colour scheme yet, but with the Knights in their first Grand Final in 25 years, we figured it was easier to ask for forgiveness than permission. Newcastle is painting the town red and blue.. so we thought we’d do our bit. Up the Knights! ⚔️ Disclaimer: No NWS buildings were harmed (or actually repainted) in the making of this post. #NewcastleKnights #NRLGrandFinal #Newcastle #NWS #NewcastleWeighingServices #UpTheKnights
-
Paul Greenwell liked thisPaul Greenwell liked thisJev is getting a lot of hype. ⚡ New 'System 1' AI models like Jev (closed-source) and Laya (open-source) are designed to make faster, lightweight decisions without needing the deeper reasoning of a larger AI model. They have some pretty cool use cases. But sometimes the fastest and most reliable solution is sometimes no AI at all. 💡 If an order can only move from: NEW → PAID → SHIPPED → DELIVERED and those transition rules are known, use deterministic code: a state machine, rules engine, validation logic or simple condition. ✅ Models like Jev and Laya become useful when the decision is fuzzy: - "Is this customer likely to churn?" - "Which team should handle this messy support ticket?" - "Does this message look like a jailbreak attempt?" A simple framework to follow: Known rule? Use code. Fuzzy, fast, bounded decision? Use a small, specialised and ideally fine-tuned decision model. Complex judgement? Use a Large Language Model (LLM) Combination of needs? Use the right mix. 'System 1' models are cool shiny new tools ✨️ But before you swap 'sprinkle AI fairy dust everywhere' hype for 'put a faster model everywhere' hype, make sure your AI selection strategy answers three questions: 1. Should we use AI at all? 2. Where does AI add value? 3. What type of AI is the right fit? At Propel Ventures, we help teams answer those questions before they start building, so they can focus investment on the AI use cases that actually improve business and customer outcomes.
-
Paul Greenwell liked thisPaul Greenwell liked thisPropel Ventures is now a Select partner in the Claude Partner Network Services Track. This recognises the work our team has been doing to put Claude into production and help people use it safely and productively. Anthropic’s Services Track assesses partners across three dimensions: certified people, production deployments and approved public customer stories. Here’s what that looks like at Propel: → 20 consultants and engineers hold current Claude certifications. → More than 20 customer engagements have been approved as Services Registrations, recognising Propel as Delivery Partner of Record. → Our published customer stories show how our AI Growth Sprint, clear governance guardrails and custom Claude plugins help firms change how they operate to becomer AI native in three weeks. We’ve also trained more than 3,000 people to work with Claude and generative AI—from boards and executive teams to engineers, analysts and front-line staff. That last number means a lot to me. Successful adoption depends on people having the skills and confidence to apply AI to their own work. Helping them get there is a core part of what we do. This relationship gives our team access to Anthropic’s partner training, product briefings and sandbox resources, strengthening what we can deliver for customers. Thank you to the Propel team who earned this recognition, and to the customers who have trusted us with their work. If you’re exploring Claude for your organisation, or looking to move from experimentation into production, I’d welcome a conversation. #Anthropic #Claude #PropelVentures More info at: https://lnkd.in/gbK5UCw5
-
Paul Greenwell liked thisPaul Greenwell liked this🚨🧭 The engineers adopted AI first. Then the product managers got found out. Two years ago every board assumed AI would come for the developers. That is not what happened. In most of the businesses we are working with right now, the pattern is the same. Engineers picked up spec-driven development in a matter of months. Close to 100% of their code is now AI-written. And the constraint in the lifecycle moved upstream, to whoever defines the problem to be solved. Product managers and business analysts are now the slowest step in the process. Most of them don't know it yet. Their engineers do. I have had a run of calls from CEOs saying the same thing: my engineers are ahead of the curve, and the other team members need to get into the AI game. The decision this puts in front of the executive is uncomfortable. Stop funding developer AI tooling in isolation. The developers are fine. Upskill the product management team, because that is where the throughput is now being lost, and it is the part of the organisation least prepared to hear it. At Propel we saw this early because we run product management and delivery for an ASX-listed client end to end. When the build got faster, the question of what to build became the whole problem. That is why our AI PDLC work starts with product strategy and discovery, not with the code.
-
Paul Greenwell liked thisPaul Greenwell liked thisEvals have become a key part of the AI PM and AI Engineer job stack. Over the last 12 months, I've spent weeks building evaluations, refining prompts and monitoring how production AI features are actually used. Evals are time-consuming up front, but pay dividends down the line. A few lessons: 🧪 Coverage beats repetitive volume Start pre-launch with synthetic data and a versioned golden dataset covering happy paths, edge cases, regressions, safety and responsible use scenarios. ⚖️ Generic eval metrics need context “Conciseness” sounds useful, but only if judged relative to the question being asked. A two-line answer might be perfect for a simple question and terrible for a complex one. Wherever possible, tie your evals to clear decisions on whether your AI feature did what it needed to do: - Did it answer the question appropriately? - Did it call the right tool? - Did it follow policy? - Did the output match expectations? And overall: Was the result Good / Okay / Bad? 🔎 Production traces are key to success Capture the traces that matter, anonymise, run automated checks and LLM judges, then surface the highest-value cases for human review. Track human vs judge disagreement as a calibration signal. Some disagreement is healthy. Too much means your judge likely needs refining. Turn useful failures into new golden cases. 🛠️ Tools that help Langfuse and Braintrust are easy-to-learn tools for testing prompts and models, and evaluating production traces. For next level benefits, consider building a lightweight, integrated review app that turns production signals into ideas, refined product backlog decisions and faster improvements. 🔁 The loop I’m using: Golden dataset → evaluate & refine → launch → production traces → evals → human review → decisions → improvements → test At Propel Ventures, we’re helping product teams adopt evolving best practices for AI evaluation, drawing on lessons across industries to build higher-quality AI features that better match their intended outcomes. #AIProductManagement #LLMOps #AgenticAI #AIEvals
-
Paul Greenwell liked thisPaul Greenwell liked thisGreat to be at the University of Melbourne Internship Showcase today representing Propel Ventures! It was a pleasure meeting so many talented Melbourne Uni students and hearing about their backgrounds, interests and career goals. We’re currently offering two paid internship pathways: • AI Engineering Bootcamp + Internship — for students with a software/engineering background, covering practical AI engineering including model APIs, RAG, MCP and enterprise AI foundations, followed by real client delivery work. • AI Enablement Internship — a more people-focused pathway, helping client teams from executives to frontline staff apply AI tools effectively in their everyday work. No coding background is required. Thanks to everyone who stopped by the Propel table today. If you’re interested in joining us, make sure you apply through the University of Melbourne internship portal or drop us a cv. Looking forward to seeing you during the interview! #PropelVentures #UniversityOfMelbourne #Internship #ArtificialIntelligence #AIEngineering #Careers
-
Paul Greenwell liked thisPaul Greenwell liked thisBased on feedback last week about my post on non AI bottlenecks to engineering efficiency, I thought I would share three non AI bottlenecks which get in the way of your AIPDLC, that I saw slowing teams down this week: 1. Code review latency. Not the review itself - the wait. A PR sitting for two days costs more than the code took to write. Fix: review SLAs, smaller PRs, and yes, AI-assisted first-pass review to shrink the human queue. 2. Handover tax. PM writes a brief, BA translates it, PO refines it, engineer reinterprets it. Every handover loses fidelity and adds days. Fix: smaller teams with shared context. We are seeing hands-on product people collapse the PM/PO/BA split entirely. 3. Meeting load and calendar fragmentation. Engineers with four hours of meetings have zero deep work blocks. Fix: protected maker time, async-first status, and killing the standing meetings nobody can justify. If you want to see your engineering team hum, we can go further than just delivering an AIPDLC - we can remove your non AI engineering blockers as well, so enable your team to really move fast.
Experience & Education
Licenses & Certifications
Volunteer Experience
-
Assistant Treasurer
KDSPA
- 6 years 2 months
Education
Recommendations received
-
LinkedIn User
3 people have recommended Paul
Join now to viewOther similar profiles
Explore more posts
-
Tauseef Khushdil
ndisync.ai ended up with roughly 64 features, but we didn’t sit down on day one with a list of 64 things to build. A lot of the product took shape while we were already building it. The client was constantly researching the NDIS industry, learning more about how things worked and bringing new requirements back to us. Sometimes that meant adding something new. Other times it meant changing something we had already built. Early on, a request could sound straightforward in a meeting. Then the development team would open the product and find that the change also touched permissions, an approval flow, data being used somewhere else, invoicing or reporting. That changed the way I look at feature requests. Today, when someone asks how long a change will take, I’m much less interested in how complicated the new screen looks. I want to know what already depends on it. That’s usually the difference between a change that takes an afternoon and one that turns into a much bigger piece of work. Building ndisync.ai taught me that software gets expensive in the connections you forget to check, not just the features you decide to add.
1 Comment -
James Pepplinkhouse
Our recent proposals have come in around 80% below what a Brisbane agency would likely quote for the same build. People often pause at that. A lower price must mean a lower standard, right? In fact the reverse is true, and the reason is testing. Well-tested software has roughly one line of test code for every line of the app itself. Written by hand, that close to doubles the work, so tests are the first thing cut when a budget gets tight. In our builds, AI agents write the code and the tests together. Every change has to pass the full test suite, and a person signs off before anything goes live. Every project is also built to our secure development standard, aligned with ISO 27001, and we attack it ourselves before launch. That is what lets us fix the price of every milestone before work starts. It isn't cheap software, it's a new business model for the agentic era. I've written up how it works, plus three questions worth asking any developer before you kick off your project: https://lnkd.in/gpiFt2KK
1 Comment
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore More