About
👋 Hey! I'm Callum - a Full Stack AI Software Engineer passionate about building…
Activity
3K followers
-
Callum Holt shared thisThe trick with AI doing TDD is that if the same agent plans, codes, and writes the tests, it just fits the tests to its own work to make you happy. You need to separate the planning, worker, and validation agents so they don't share context. I find software engineering to be in a super interesting spot at the moment. At most of the events i've been presenting at recently, ive been talking about how AI is incredible at the execution stage of software development, but its still incredibly important for software engineers to not give up on the thinking stage. Spending the time thinking about architecture and data flow and how systems connect together and potential bottle necks is still incredibly important. The testing component is where it gets interesting. If you just tell the AI thread to develop the plan, execute and then do 'tdd' it absoltuely works, but it tends to fit the tests to the plan to ensure that it succeeds. The bottle neck then becomes the human in the loop doing the testing of software that is produced to make sure that it meets spec. OpenAI discussed this in their "Symphony" blog post a couple weeks ago, which also references their earlier "Harness engineering" post from feb. The next layer i'm finding is building an automated e2e testing system that the ai can control to validate its own work and confirm that other bugs haven't been introduced, in the way a human #QA would test. But only if the planning, worker and validation agents are kept separate - otherwise the worker just builds tests that validate its own work.
-
Callum Holt shared thisLast week I ran a workshop for Swinburne Innovation Studio. 120+ founders on Zoom. The topic was #AI for #sales and #marketing. But the thing that got the biggest reaction wasn't a slide. It was when I opened my terminal and showed them how you get agents to actually do real work, not just perform question answer queries. Over the last few months I've been expanding how I use #Claude Code. It started as a coding tool. Now I treat it as an agentic operating system - building reusable multi-step programs for the actual work of running a business. I use claude code for most of them, but the same patterns apply to #codex and OpenCode as well. Here's what I showed the room: → /build-audience - AI scrapes Google Maps for 150 businesses, finds owner names, validates emails, researches niche pain points across Reddit and forums, then writes 100 personalised cold emails. One command. Under 30 minutes. → /generate-seo-content - takes customer voice data scraped from forums and reviews, then generates blog posts structured so ChatGPT, Perplexity and Google AI Overviews can actually cite them. Answer-first sections. Stats every 150 words. FAQ with schema. → /deep-research - spawns parallel agents across 20-30 queries. Multiple rounds with gap detection. Produces a research dossier scored by confidence. These aren't prompts. They're programs. Each step feeds the next. The question I keep getting asked: "How long did this take to build?" Honest answer - the first version of each took a weekend. Getting them reliable took months. But now they run in minutes and get better every time. The thing I kept coming back to in the session: Most people are still using AI as a slow Google search. Question in, answer out. That's level zero. The real unlock is getting AI to actually execute. Research, enrich, write, validate, send - not just advise. I think. AI executes. Repeat. One of the last questions was about introducing AI into a business built on human relationships. My answer: don't automate the relationships. Automate everything that prevents you from doing more of them. Huge thanks to Swinburne University of Technology Innovation Studio and for Audrey Jean-baptiste and Jennifer Bevis for the platform to teach people about agents, marketing and sales! If you're a founder still prompting one question at a time - you're leaving 90% of the value on the table. I'm super excited to be running more of these types of workshops in the future! Reach out if you want the recording of this and ill send you a link 🔥
-
Callum Holt posted thisThe biggest unlock in my engineering workflow this year hasn't been a smarter model. It was closing the loop. Most people are running Claude as an open loop. Prompt in, code out, human checks it, human runs it, human reports back what broke. The model is doing 10% of the work and you're doing the other 90% just being the messenger. A closed loop is different. The model triggers an action, sees the result, and uses that result to decide the next move. No human in the middle. The feedback is the fuel. Two examples from this week. Testing new pricing. Claude had the Stripe sandbox CLI, an MCP server on the database, and Claude in Chrome for driving the UI. It created a test product, pushed it through checkout, queried the DB to confirm the subscription landed, then adjusted when something was off. PR reviews. Claude tails the Claude Code PR review on GitHub, reads the comments on its own PR, implements the fixes, then updates my local PR review agent with the issues that got flagged so it catches them next time. Then pushes back to GitHub. The system gets smarter every cycle. I went from prompt operator to reviewer. The interesting bit isn't any single tool. It's that the moment you give the model a way to observe what it just did, it stops being a chatbot and starts being a system. Most workflows are one tool short of being a closed loop. Find the missing one.
-
Callum Holt posted thisOpus 4.7 shipped this morning. The benchmarks are impressive. The breaking changes are more interesting. Temperature, top_p, top_k - gone. Set any of them to a non-default value and the API returns a 400 error. Extended thinking budgets - gone. Adaptive thinking is the only option now, and it's off by default. Thinking content is hidden unless you explicitly opt back in. If you're running production code against Claude, that's three things that will break when you upgrade. Not deprecation warnings. Hard errors. The actual capability improvements are real though. Tool errors dropped by two-thirds. SWE-bench jumped from 53% to 64%. Vision went from 1.15 megapixels to 3.75. The model follows instructions more literally and spawns fewer subagents by default. The tokeniser is interesting. Same pricing - five dollars in, twenty-five out. New tokeniser uses up to 35% more tokens for the same input. But usage limits have also apparently been increased to make up for the extra tokeniser tokens. Before we know it, the next labs will release their next competitors as well!
-
Callum Holt shared this"You changed my life." One attendee's words after deploying his first app to a live URL. buildDay ran over Easter Saturday at Stone & Chalk, Melbourne. 8 non-developers. 6 hours. Built and deployed a full-stack app from scratch using AI. The build: an outreach generator. Feed in a prospect's name, company, role, and pain points. It hits an AI model via OpenRouter, returns a personalised cold email and LinkedIn message. Saves to a Supabase database. Deployed live on Vercel. Cursor's agent mode did the coding. Nobody wrote a line manually. Plain English in, working software out. Three lessons that landed hardest: Commit and push after every working change. Your last commit is your life raft when things break. Copy the error, paste it to the agent, let it fix. Debugging is the real skill - not building. AI is fantastic for execution. Dangerous as a replacement for your own thinking. Another attendee had tried building apps for months. YouTube, Udemy, gave up every time. His take: "I've been trying to find something like this for ages." Thanks to Cursor, Supabase, Vercel, OpenRouter, Stone & Chalk Melbourne and MLAI for making it possible. well done Timothy, Haoxuan, Hinc, Peter, Kem and Isaiah! You guys did fantastic and i cant wait to support you in the future! Next buildDay releasing soon. DM me or drop a comment if you want in.
-
Callum Holt shared thisSuper excited to be teaming up with the Swinburne Innovation Studio to run this masterclass at the end of April! As a senior software engineer, consultancy operator and AI practitioner, I'll be breaking down: → The three different levels of AI products → Critical traps founders fall into when using AI → Powerful sales and marketing strategies I've been using for both inbound and outbound On the inbound side - I grew my SEO and AEO from 0 to over 2,500 impressions in a single week. On the outbound side - I've been building hyper-personalised B2B outreach pipelines using Claude Code CLI. Whether you're building an AI product or just trying to figure out where AI fits in your go-to-market, this one's worth an hour of your time. Free. Online. April 30.Callum Holt shared thisNot sure where AI fits in your startup? You're not alone. Most founders are using AI for product. But very few are using it to punch above their weight in sales and marketing. In this April masterclass, we're joined by Callum Holt, software engineer, consultancy operator and AI practitioner, to break down exactly how founders can change that. What you’ll learn: ✅ The three levels of AI products ✅ How to use AI to accelerate your Go-to-Market across sales and marketing ✅ The critical thinking traps founders keep falling into ✅ And just as importantly, when AI isn't the answer Event details: 📅 Thursday, 30 April 2026 ⏰ 2-3pm (AEST) 📍 Online via Zoom (Free) Whether you're building an AI product or just using AI tools to grow, this session gives you a practical framework to make smarter decisions. Secure your seat today ➡️ https://swi.nu/tz9nw
-
Callum Holt shared thisI've been experimenting with using AI agents to build programmatic SEO and AEO at scale. About 10 days ago, I used Claude Code to produce 4,322 targeted pages for my site in a single day. Not thin content. Not template swaps with a city name changed. Pages built on real research. Here's the process: 1. Pain point research first, production second Before generating a single page, I scraped and analysed hundreds of sources across the web - Reddit threads, forums, competitor reviews, search queries - to understand what people actually search for and what problems they're trying to solve. This became the foundation for every page. 2. Built data models for every dimension I structured the content around 50 industries, 104 locations, 6 service types, 131 app types, 200 tool comparisons, and a 504-term glossary. Each combination creates a unique, specific page that matches a real search intent. 3. Phased production with audits between each phase Phase 1: Core matrix pages (service x industry, service x location). Phase 2: Expanded comparisons, glossary, and industry cross-references. Phase 3: Location x service and build x location combinations. After each phase, I ran an SEO audit pass before moving on - fixing content uniqueness issues, schema markup, and internal cross-linking. 4. Researched what makes AI content "slop" - then built a system to avoid it This was the part most people skip. I compiled research from eight different sources on what makes AI-generated content detectable - the overused words, the em dash addiction, the "in today's fast-paced world" openings, the corporate filler. Then I built that research into a copy editing skill that runs as part of the content pipeline. Every page gets checked against a detection reference of ~200 known AI writing patterns and rewritten to sound like a human actually wrote it. The results after 10 days: - 15,500 impressions in Google Search Console - Impressions went from near zero to 2,400/day - Average position 20.3 and climbing - Clear hockey stick growth starting the day the pages went live The takeaway: the hard part of programmatic SEO isn't production anymore. AI agents have made that trivially fast. The hard part is the research that goes in before you write anything, and the quality systems that stop your content from reading like every other AI-generated page on the internet.
-
Callum Holt shared thisRecently I've been doing a heap of #agentic #SEO and #GEO (generative engine optimisation), and one of the tricky things is not having a clean way to quantitatively measure how a domain is perceived by AI. When someone searches on ChatGPT, Gemini, or Perplexity for a solution that you and your competitors both offer - does the AI mention you? And if so, how prominently? There was no straightforward way to answer that. So I built one. It lets you enter a URL, generates realistic queries someone might ask an AI about your industry, runs them against AI providers, and checks whether your site gets cited in the responses. You get a clear visibility score - a quantitative measurement you can actually track over time. I built it to solve my own problem. Traditional SEO tools have existed forever, but nothing really gives you a read on how visible you are to AI assistants specifically. And with more people skipping Google entirely and going straight to AI, that felt like a pretty significant blind spot. I figured if I needed this tool, other people probably do too. As an engineer, I gave it an extremely technical name - Does AI Rank Me. Feel free to check it out, have a play, and let me know what you think.
Callum Holt commented on a post
3d
love it mate! Pretty much identical set up to mine! I do find though with Amphetamine that my laptop seems to need significantly better cooling than normal? have you noticed increased heating w amphetamine? might be just me on my macbook air without a fan....
Callum Holt replied to a comment
4d
Sal Samani that's interesting! Did you see any teams at colosseum working on POS systems W crypto native reporting?
It seems like a no brainer for merchants to adopt because cost of settlement is so much cheaper compared to mainstream rails. But obviously it's a bit of chicken and the egg because there also needs to be consumer demand. Although I'm sure that could be generated by passing through the transaction cost savings as discounts to customers or rewards programs etc.
I dunno, it's an area that I'm pretty interested in so I'm following closely.
Callum Holt replied to a comment
4d
Sal Samani what do you think would have to happen to be able to do that? Like as far as I'm aware technically merchants could already accept payments from customers with the settlement layer being on Solana? Even through centralised services like stripe or decentralized through platforms like Jupiter that offer onchain merchant services.
What do you think?
-
Callum Holt reacted on thisCallum Holt reacted on thisWe’re proud to announce our new partnership with MLAI, bringing together Australia’s largest innovation community and one of the country’s most engaged grassroots AI and machine learning networks. This collaboration is built on a shared belief: Australia’s AI future will be shaped by practical implementation, technical capability and community-led innovation. By combining MLAI’s deep technical expertise and hands-on AI community with our founder ecosystem, commercialisation support and national innovation network, we’re creating more opportunities for: ✅ AI founders and startups ✅ Engineers, researchers and roboticists ✅ Students and emerging builders ✅ SMEs and corporate innovation teams We're kicking things off by supporting AI Week Melbourne (1–7 June) – featuring conferences, workshops, hackathons, robotics events, community meetups and hands-on experiences designed to make AI more accessible, practical and collaborative. And this is just the beginning. MLAI will also establish a permanent presence within our Melbourne hub, creating an ongoing home for Australia’s AI community to connect, learn and build together. The future of Australian AI will be built by people who experiment, collaborate and create. We’re excited to help bring this community together. Read more: https://lnkd.in/eFZDnn4v Stone & Chalk MLAI Dr Sam D.
-
Callum Holt reacted on thisCallum Holt reacted on thisIf you are an AI engineer, a technical founder, or someone building something in stealth who wants to spend 26 hours in a house with Antler and the best builders in Australia, this is for you. Here is what every selected founder gets: - All expenses paid. House, food, transport, everything covered. - $1K in Azure credits from Microsoft. - 1 year free on Vercel - Free legal consultation with Allied Legal - And if you win, $25K in prizes from Microsoft. Build Island. Phillip Island. June 4 to 5. Build Lab. Sydney. June 11 to 12. Apply here: https://lnkd.in/gxc2XPBy Tag someone who should be in this room.
-
Callum Holt reacted on thisCallum Holt reacted on thisI moved to Australia in 2004 with not very much, and over the years I have been fortunate enough to get on and make something of it. This budget kills aspiration and the very get-up-and-go spirit that drove me and built Australia into the country it is today. My parents fled communism in 1956 and I was brought up to work hard and not take handouts. I don't support big government. I have seen first hand what communism does to a country, through regular trips to Hungary with my parents during communist times. It is a complete destruction of productivity. There is nothing in the budget to drive productivity, quite the opposite. A lot of the discussion around this budget misses the point. All the government was supposed to do was help first home buyers get into the housing market. It's quite telling that the budget fundamentally does very little to achieve that. There is something much bigger happening. It's a fundamental shift toward a welfare state, which requires an ever-growing need to raise taxes. If we accept the budget as proposed, next years budget will include inheritance tax and more. I firmly reject the proposed tax changes and have made a formal submission to the Senate inquiry on the proposed CGT reforms. If you feel the same way, do the same. Click the link below. It doesn't need to be long. https://lnkd.in/eu3zVCxb Or copy and paste this: On 13 May 2026, the Senate agreed that the provisions of all time critical bills introduced into the House of Representatives after 14 May 2026 and up to and including 4 June 2026,
-
Callum Holt liked thisCallum Holt liked thisif i was pablo escobar, here are the first 3 things i'd automate. pablo escobar ran a $30B empire on notebooks and payphones. no CRM. no automated follow-up. no real-time financial visibility. 1. Cash reconciliation at his peak, escobar was making so much money he was spending $2,500 a month just on rubber bands to bundle the notes. he literally wrote off 10% of his cash every year to rats eating it in storage because he couldn't track it fast enough to move it. an agent reconciles every transaction in real time. flags discrepancies the moment they appear. 2. Supplier and logistics tracking the whole operation ran on manual coordination across dozens of suppliers, routes, and middlemen. one missed handoff, one miscommunication, and the whole chain breaks. an agent monitors every supplier, every delivery window, every threshold. when stock hits a certain level, the order places itself. 3. Knowledge base management escobar's biggest problem wasn't the DEA. it was his own people not knowing what they were doing. missed payments. conflicting instructions. promises made that nobody wrote down. an agent tracks every commitment, every conversation, every outstanding obligation across the whole operation. nothing gets forgotten. even a $30B cartel would have run leaner with AI.
-
Callum Holt liked thisCallum Holt liked this"What's your AI strategy?" It's a question coming at founders from customers and investors that I'm hearing a lot of uncertain answers to right now. This isn't a build question. It's a positioning question, and you need an answer regardless of what's in your stack. If there's no AI in your product, you need a clear, defensible reason why. If you do have an AI component, where it sits in your messaging becomes a strategic decision. Headline feature? Quiet capability? Buried in the stack? The right answer depends on who you're selling to, what they care about, and what your competitors are already claiming. I'm running a 60-minute masterclass on this next Monday, 25 May at 12pm AEST on Zoom. $49.50 a seat. We'll work through five questions that help you decide: • whether AI should be central to your positioning • how to talk about it credibly • where founders accidentally create buyer distrust • when AI helps differentiation and when it just creates noise • what answer you want your team giving when people ask about your AI strategy You'll apply the framework directly to your own product, buyer, and market positioning and leave with an answer you can actually defend to customers, investors, and your team. If you want to refine your AI positioning, or haven't properly worked through it yet, this is the session to come to. I'd love to see you (virtually) there! Link in the comments.
-
Callum Holt liked thisCallum Holt liked thisEvery day, Uber runs a global real-time marketplace at massive scale. Using OpenAI models and Realtime APIs, Uber is building AI-powered products that help drivers make smarter earning decisions and help riders book faster. Uber Assistant turns marketplace signals like trends and heatmaps into simple guidance, while new voice experiences let riders request trips using natural speech. Uber has expanded Uber Assistant beta experiences to hundreds of thousands of U.S. drivers - helping better support drivers on the platform. Learn more at the link in the comments below.
-
Callum Holt liked thisCallum Holt liked thisso we hit 3.1k users yesterday 🎉 (up again from 2.8k peak last week) we've been growing consistently for the last 2 months, and our new study room feature made a big impact! - we help students study together directly on polarbear 🙌 also, i was privileged to be one of the coaches for the recent Hack48 hackathon which was super cool -> congrats to Chris Yoo and the team, they're building something genuinely exciting anyways, update over, back to work
-
Callum Holt reacted on thisSee yall next week. Hosted and coordinated by AI Engineer Web Directions Stone & Chalk and MLAI LOTS of juicy giveaways and slick af events in store... Check the FULL lineup of events and side events int he comments John Allsopp Jun Kai Chang - Luc Sonia Kaurah Dr Anu G 🩺 Yana Lin Christopher Moss Alisa Belova Shan Yang Juan David Bernal Pegah Khaleghi Angie Simmons Anjali Singh Kaey-Lib Tan Jessica Cheney Grace Moss Aaron Zachariah David Gilmore Gabriel Chua Doron Bahar Renee Dvir Omer Shai Julian Featherston James Kahn, PhD, Bill Lilley Hao Wang
-
Callum Holt reacted on thisCallum Holt reacted on this🚀 Loyalmate v3.2.0 is now live! One of our most requested upgrades is finally here: Location-based search is now enabled 📍 You can now discover Loyalmate businesses near you directly in the app. Why did we build this? As more users started discovering Loyalmate through content and word of mouth, we realised something important: People shouldn’t download the app and wonder: “Where can I actually use this?” Now, slowly but surely, we’ll be updating more locations near you so you can explore and experience Loyalmate at cafes and businesses around your area. But this is an exciting step towards making loyalty more seamless and discoverable. And yes… this is only the beginning 👀 Want to check it out? Download Loyalmate and explore businesses near you: https://loyalmate.app Would love your feedback on the new update 🙂 Stay tuned for the next release. We’re building something exciting. #buildinpublic #startup #founderjourney #productlaunch #customerretention #loyaltyrewards #startuplife #saas #australianstartups #productdevelopment #earlystage
Experience & Education
-
Shopfront
View Callum’s full experience
See their title, tenure and more.
or
Licenses & Certifications
Honors & Awards
-
Australian Institute of Fitness Class Champion
Australian Institute of Fitness
The AIF Class Champion Award is a prestigious honor recognizing students who demonstrate exceptional dedication and determination in achieving their fitness goals. Qualifications include:
Master Trainer Eligibility: Champions maintain eligibility for Master Trainer certification, avoiding three strikes in one category or five strikes across all.
Going Above and Beyond: Champions excel by dedicating extra time to independent study, honing practical skills, and assisting fellow…The AIF Class Champion Award is a prestigious honor recognizing students who demonstrate exceptional dedication and determination in achieving their fitness goals. Qualifications include:
Master Trainer Eligibility: Champions maintain eligibility for Master Trainer certification, avoiding three strikes in one category or five strikes across all.
Going Above and Beyond: Champions excel by dedicating extra time to independent study, honing practical skills, and assisting fellow students.
Diverse Experience: Champions learn from Coaches and industry professionals, gaining valuable client perspective.
Proactive Feedback Seeking: Champions actively seek input for personal growth from Coaches, mentors, peers, and clients.
Building Connections: Champions radiate positive energy, making them a magnet for inspiration and motivation.
Winning grants industry recognition, enhances employability, and includes a prize from Fitness First. It's a powerful addition to your resume, setting you apart in the competitive field of Personal Training. -
ADF Long Tan Youth Leadership & Teamwork Award
Australian Defence Force
The ADF Long Tan Youth Leadership and Teamwork Awards started in 2006 to recognise students who demonstrate leadership and teamwork within both the school and the broader local community. At the same time, they recognise those who display strong values, such as doing one’s best, respect for others and “mateship”, characteristics that are integral to Australian society.
The awards are an important initiative by the Australian Government in conjunction with the Australian Defence Force…The ADF Long Tan Youth Leadership and Teamwork Awards started in 2006 to recognise students who demonstrate leadership and teamwork within both the school and the broader local community. At the same time, they recognise those who display strong values, such as doing one’s best, respect for others and “mateship”, characteristics that are integral to Australian society.
The awards are an important initiative by the Australian Government in conjunction with the Australian Defence Force designed to encourage senior students – tomorrow’s leaders – to actively participate in the life of their schools and local communities. -
La Trobe University Award for Outstanding Achievements
La Trode University
Recommendations received
2 people have recommended Callum
Join now to viewOther similar profiles
Explore more posts
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