About
I'm a go-to-market systems and operations leader with 15+ years experience building…
Articles by Justin
Activity
10K followers
-
Justin Norris posted thisI think Luna xhigh (in Codex) is my new favorite LLM for just GSD. Fast and efficient, no posturing or jargon, and so very competent. The writing, while a bit austere, actually sounds like how a human communicates, not stuffed with slop. And you can use it non-stop on a $20 ChatGPT subscription without worrying about limits.
-
Justin Norris shared thisThe conversation around AI is slowly shifting from "look at all these cool things I can do" to "here is the economic value AI is bringing to the business." Don't get me wrong - many AI capabilities are emerging every day, and they are amazing. But what was novelty yesterday is table stakes today. And cost-efficiency starts to matter a lot. For many ops and AI transformation teams who have been laser focused just on adoption, this will be a developing skill set. Here are a few things you should add to your list if not already on there: DESIGNING FOR EFFICIENCY Every token and tool call has a cost. Your agent's massive system prompt gets passed to the model every turn. Good design pays dividends across every run, no matter what model you use. Topics to think about: Skills, progressive disclosure, sub-agent structures, prompt caching, using cheap classifiers before expensive agentic loops, etc. INSTRUMENTING FOR COST OBSERVABILITY It's a helpless feeling to see your API bill climb without knowing exactly why. If your agents/workflows aren't instrumented properly, you're guessing. You need to log costs for each model call BY CALL SITE. I did this for one agent and realized some crazy things, like - I was paying to load a lengthy system prompt to do a simple classification - my RAG searches were returning way more chunks than necessary - I had 20 redundant tool descriptions All very costly mistakes that stuck around for every conversation turn. CHOOSING THE RIGHT MODELS People are so focused on which model is "best" without asking "what sort of intelligence do I need for this task". It's important to go beyond the rate card and the public benchmarks and perform head-to-head bake-offs with your real-world workflows to see the true cost-per-task. A "cheaper" model can be more expensive if it uses more tool calls and spins its wheels to get the same output. This week I did that bake-off, taking one of our most-used agents and running a real-world prompt through 11 different models. The results were eye-opening. The cheapest models were not the cheapest to do the task. Open-weight models that I thought would perform well bombed, in some cases, not in others. Also my prompt - hill-climbed extensively on Anthropic models - did not translate cleanly to other families. CONTINUOUS OPTIMIZATION AI economics change constantly. We're in a state of flux. New models come in all the time, the labs make changes that can have massive economic impact. So you need to revisit your assumptions pretty much weekly. E.g., OpenAI cut the cost of GPT-5.6 Luna today by 80%. That's the equivalent of GPT-5.4 xhigh intelligence (state of the art just a few months ago) at a fraction of the cost. --- Anyways, just a friendly note to add "AI Cost Engineer" to the list of hats you now need to wear.
-
Justin Norris shared thisEvals are standard practice for AI Engineers, but from what I've observed don’t seem that common for internal AI tooling yet. This is a big gap and fortunately isn't that hard to fix. Evals are a core part of professionalizing the internal AI stack. Just as mature Systems teams have automated QA and regression testing for Salesforce, Marketo, and other core systems, your agents should have automated evals to test their behavior. The other benefit is that when one agent can build another agent and test its own work, you can automate much more of your development lifecycle.How to build a Claude plugin marketplace with evalsHow to build a Claude plugin marketplace with evals
-
Justin Norris shared thisI feel like a lot of AI posting right now is "here's how you get good at AI" or "look at this cool thing I built." At the individual level, that makes perfect sense. But at the company level, it's not enough to have a few AI wizards. The bigger question is how you identify the most impactful capabilities and distribute them through the org as quickly as possible. Example: it's all well and good to have an ace salesperson who uses AI to save 10 hours a week and increase their close rate 20%....but the impact is limited unless you can replicate that across your entire field. It's a lot like traditional L&D or enablement. Except now you can insert validated skills directly into the brain of a person's AI assistant, which is pretty cool. It's Neo learning Kung Fu in real time. I haven't seen much discussion about this part of the work, so wanted to formalize my thoughts a a bit more in this post.
-
Justin Norris posted thisOne of the most difficult questions right now is "what value is AI actually bringing to your organization?" This is like the 2026 version of marketing attribution. You know some things are working, others not. People want scientific answers but it's messy and difficult to prove. Also there's no getting away from it. In my initial thoughts there are a few different ways to answer this question. In ascending order of difficulty. 1) Task-based efficiency: easy. Task A used to take us one hour (measurable) and now takes us 5 minutes. The delta x # of task runs in the year = capacity created. BUT this is only valuable if you can redeploy that capacity productively (i.e., it's not wasted) or you save on costs. If that's true, and the benefit exceeds the cost of the system, you have ROI. 2) Unit volume efficiency: also easy to measure. We used to produce x widgets for a given input. Now we produce x+y for the same input at equal or better quality. The surplus is likely due to AI. Good for things like support tickets. BUT this only works if you don't lose quality. Otherwise the volume can be a vanity metric or actively harm the business (e.g., more tickets closed but less happy customers). 3) Surplus productivity: this starts getting fuzzy. How do you measure "extra work I did because AI makes me faster at everything" in the absence of a productivity metric that makes comparison possible? E.g., I know for a fact I'm just getting more things done thanks to AI. But I don't have an easy way to prove it. A lagging indicator could be absorbing more growth without more headcount, e.g., a rep managing a larger book of business than before. 4) Quality improvement: to me this is the hardest but also the most valuable and strategic. It’s AI helping a person work on the right things, a manager provide better coaching, a seller execute better, a leader have a better strategy. The return here can be vastly more valuable, but attributing improvement to AI is difficult and expensive. In some cases it can be possible to measure via a KPI such as win rate. But even in that case, you would need to stage it like an incrementality test - a randomized sample of reps using a given AI workflow, a holdout without, compare performance over time. I don't think many companies are doing that. 5) Capability creation: work that wasn't happening before because it was too expensive or operationally impractical. E.g., we used to hand review a few dozen closed opportunities or prospect calls. Now we use AI to systematically review hundreds or thousands. This value can be incredibly strategic but is difficult to attribute. We need a way of drawing a line between "I got this insight" and "I achieved this outcome." Would love to know other methods people are experimenting with to gauge success of their AI programs. Ultimately I suspect there’s still a fair bit of extrapolation and finger-in-the-wind measurement here, more than most companies would care to admit.
-
Justin Norris posted thisNeed help from my network. I'm working on refining the AI tech stack org-wide at my company, and I want to understand what other companies are actually doing - things like: - SURFACE: Where do employees access AI? (E.g., Claude/ChatGPT app, something homegrown, other tool?) - MODELS: One model family or multiple? Are you considering moving towards open weight / mixture of model approaches? - KNOWLEDGE/TOOLS: How is access governed? (E.g., per system MCPs with user-level auth, central database with permission-based retrieval, etc.) - INNOVATION: How are role-level innovations identified and shared within the org? I'm thinking here less about "what cool agents have you built" and more about "what does your company-wide AI stack look like" (focused more on knowledge workers outside of ENG). If you can share a quick summary as a comment that would really help me - DM also fine if you'd prefer. Happy to share back a summary of what I learn.
-
Justin Norris shared thisA few months ago I wrote an article about the limits of the current AI tool ecosystem for knowledge workers in general. Claude Code and Codex emerged from engineering workflows, and the ergonomics are designed around coding. My thesis was that simply wrapping them in friendlier UIs (eg, Cowork) doesn’t solve the fundamental problem that knowledge work happens in collaborative surfaces (Docs, wikis, multiplayer chat, etc) and not on a user’s local machine. AI needs to be a first class citizen in those collaborative spaces in order to be tightly integrated into a broader range of knowledge workflows. This release of “Claude Tag” from Anthropic is interesting to me as it shows how that might actually happen. There’s a few paradigm shifts wrapped inside this feature 1) accessibility - you can tag AI in a collaborative space, starting with Slack, but one can easily see how it might extend to cloud documents, knowledge bases, project management systems, or any other cloud system. This alone solves a lot of copy paste work 2) multiplayer - I already find my team is doing a lot of “my Claude talks to your Claude” type of work with us in the middle. Having a single shared agent that is context-specific as opposed to owned by one user allows everyone in that space to interact with it. No having multiple people running side missions with their own LLM thread 3) persistent context and proactivity - it’s unclear how it will work exactly, but the agent will retain context of the space it’s in and even be able to act proactively in an OpenClaw like fashion. Now you don’t have to explain the same context (or recreate context stores locally) and in theory you may not need to wire up external triggers to get AI to act. Lots of unknowns - how are these agents governed and managed? How do you control what they retain? How does a tagged Claude interact with a local Claude if at all? But still you can see the bigger shape of things to come if you stand back.
-
Justin Norris reposted thisJustin Norris reposted thisI ran an AI agent on Gong G2 reviews to map the triggers Gong's buyers describe that the category does not write to. This edition builds on work by Justin Norris who documented an AI workflow for extracting Category Entry Points (CEPs) from discovery calls. I adapted it for teams without a full ops stack — and ran it on G2 reviews instead. CEPs are the situations that send buyers into market before they know what the solution looks like. The content that meets them at that moment determines who gets considered. Most category content arrives too late. Three gaps stood out in the sample analysis. - Cross-functional buyers the category does not acknowledge Engineers. Product designers. Customer success managers. All using conversation intelligence to solve problems the category does not name in its messaging. They are leaving detailed trigger descriptions in public reviews. The category does not acknowledge them as buyers. - Account handover A buyer inherits accounts. No call history. No record of what was promised. Three reviews described this as the moment they entered the market. The category does not write to this trigger. The vendor who does owns it by default. -Rep-led self-coaching The category frames coaching as something managers do to reps. Eight reviews described something different: individual contributors reviewing their own calls, comparing talk ratios, improving without manager involvement. Different trigger. Different content brief. The methodological lesson matters as much as the findings. At 20 reviews, marketing voice gap ranked second. At 90, it fell to fifth. A small sample creates hypotheses that look like findings. Sample size is a content strategy problem, not just a statistical one. The workflow: a main orchestrator agent ingests reviews, dispatches sub-agents in parallel to classify each against a CEP taxonomy, then aggregates into a ranked report. Each sub-agent returns structured JSON. The main agent never reads raw text. Hallucination risk stays low. Edition 06 of The Content Insider walks through the full architecture, the complete prompt ready to run in Claude Code, and what the findings mean for content strategy. The gap is the brief.What G2 reviews can tell you about the content your category is not writing.What G2 reviews can tell you about the content your category is not writing.
-
Justin Norris reposted thisJustin Norris reposted thisI just shipped the biggest and weirdest content project of my career: A 4-part, 40,000-word podcast series on marketing architecture, built from nearly 50 expert voices across Humans of Martech episodes. And I turned the whole thing into an RPG-style dungeon crawl with bosses. Loot drops. Skills. Equipment. Achievements. Cuz you know, marketing architecture is becoming one of the most important conversations in GTM. As AI tools and agents take over our stacks, marketing and sales teams need clearer systems for data, governance, measurement, ownership, context, semantics, decision provenance and a bunch of other stuff. I thought the best way to explain all of that was with boss fights and loot. Obviously. So this series turns the work of marketing architecture into a progression system: Each boss represents a real org problem (like source of truth and context rot) and each skill/weapon helps you navigate the stack, the workflows, the data, and the politics required to defeat the boss. I’m honestly really proud of this one, easily the most ambitious thing I’ve built for the show so far, and it pushed me waaay beyond the usual podcast-to-blog-to-newsletter production loop. • Part 1, the fall of CRM gravity: https://lnkd.in/gjmre7am • Part 2, the eye of context: https://lnkd.in/gFyKF3KH • Part 3, the correlation masquerade: https://lnkd.in/gDGsvu_D • Part 4, the dispatch tower: https://lnkd.in/gGSfUdBq Welcome to the dungeon, martech crawler. -- Massive shoutout to all the voices featured in part 1: Meg Gowell, Head of Marketing at Elly, ex-Typeform István Mészáros, Founder and CEO at Mitzu David Joosten, Co-Founder at GrowthLoop Kevin White, Head of Marketing at Scrunch, ex-Segment John Saunders, SVP Innovation at Power Digital Marketing Lourenço Mello, Director of Product Marketing at Snowflake Erin Foxworthy, Foxworthy, Global Industry GTM Lead at Snowflake Daniel Lambert, Head of GTM Engineering at LangChain, ex-dbt Hope Barrett, Sr Director, Marketing Tech at Weather, ex-SoundCloud Blair Bendel, Sr VP of Marketing at Foxwoods Resort Casino Sarah Krasnik Bedell, Founding Growth Marketer at Railway and Scott Brinker, you know him already lol -- Thank you to our sponsors for supporting the podcast: 📧 MoEngage: Customer engagement platform that executes cross-channel campaigns and automates personalized experiences based on behavior. 🎨 Knak: Go from idea to on-brand email and landing pages in minutes, using AI where it actually matters. 🔁 GrowthLoop: The agentic, composable CDP that drives compound growth by uniting your cloud data + AI into one marketing engine. 🔌 GrowthBench: Twilio’s top-tier consulting partner, turning your Twilio investment into a customer engagement engine. -- PS: no you don’t need to have read any of the Dungeon Crawler Carl books to enjoy this. PPS: yes you will have a fun time if you have read DCC haha.
Justin Norris replied to a comment
1d
Nadia good analogy!
Justin Norris replied to a comment
1d
Peter yeah that’s a very good question. I suppose the honest answer is that it’s far from scientific today. Especially as I don’t have a way to observe usage details within license based copilots like Claude.
So for me, today, optimization is focused on the second category, where we have a defined workflow or agent and can observe cost across runs.
Justin Norris commented on a post
1d
Thank you for the creepiest LinkedIn notification I've ever woken up to!
But seriously I would never choose you for competition, you're too good
Congrats on the new gig and an amazing run at Uptempo. You were so impactful, which I saw first hand participating in the MOPS townhalls you championed.
Also I should say I still think about our podcast episode regularly. Your feedback about giving advice quickly is something I think about often (and did today).
-
Justin Norris liked thisJustin Norris liked thisCan't believe it's already week 11 of the weekly GTM AI use case spotlight from inside Zapier. Real builds, real people. This week: Donna Fung, Brand Studio Donna is a Sr. Brand Designer based in Ontario. She's the kind of designer who describes shipping a custom .tsx component in Framer with zero fanfare. This week's build: the Zapconnect 2026 agenda page, designed and shipped in Cursor. Tools: Cursor Zapier MCP Figma MCP Framer + Coda The before: every year the agenda page for our annual conference got rebuilt from scratch. We've gone from hacking 3-column blocks, then engineers coding it from scratch over weeks, then last year embedding a third-party agenda widget that had all the right functionality but was so finicky the team had to manually extend the page height every time a session was added. The after: the entire page, designed and built by one brand designer in Cursor. Pulls from an approved Coda doc as the single source of truth and outputs filtering, search, localized time, tagging, add-to-calendar, and speaker views. Refreshes daily via Zapier MCP → Google Sheet → JSON → Framer. Live at https://lnkd.in/es7u7DAM. 10 hours by Donna vs. ~2 weeks of engineering. The thing Donna built in that I think is most underrated: Figma MCP. Cursor pulls icons, components, and brand variables straight from Zapier's design system. The AI is constrained to the design system by default. It cannot go rogue on design. The design system is the thing making AI usable here, not the other way around. What I think this means for brand designers and marketers: Brand design is becoming infrastructure. In an AI-authored world, the design system is the constraint layer that makes AI output publishable. Mature system, ship-quality output. No system, generic AI-slop marketing pages. The ceiling on what one designer can ship goes way up. Donna designed, built, wired the content pipeline, and shipped. That used to be four steps by 4 different people. Super curious, designers and brand marketers doing this, how mature does your design system need to be before AI tooling actually starts paying off? (And if you haven't registered for ZapConnect yet: zapier.com/zapconnect. Donna's page will greet you.)
-
Justin Norris liked thisJustin Norris liked thisIt's time for everyone to stop using AI commenting tools AI viral post generators AI LinkedIn DM bots AI SDRs for outbound Rented LinkedIn profiles Purchased LinkedIn profiles Chrome extension LinkedIn tools You can do so much already. You don't need these hacky things. One well placed DM is gold, but a thousand spam DMs are actively harming you. One insightful post works wonders, but ten AI slop posts will never see the light of day. A thoughtful comment on a Dream Account CEO's post will open doors. But a single tone deaf AI spam comment on that same CEO's post, gets you blocked forever. You can move from one automation hack to the next, or you can sit down and do the emotional labor of listening, caring about, and actually serving your customers. Time to stop short term gimmicks once and for all. Do it today.
-
Justin Norris reacted on thisJustin Norris reacted on thisFour years ago today, I lost my husband, Anthony. He was 47. To many, he was a leader, mentor, and friend—someone people were naturally drawn to. In his work, he was always highly successful in what ever field he did. To our children and me, he was our safe place. Anthony experienced bullying and a breakdown at work. In the year that followed, he fought a battle most of us never saw. His death by suicide changed our lives forever and led me to a purpose I never expected: speaking openly about mental health, grief, and the importance of connection. In workplaces and communities, we often celebrate resilience and success, yet people can be carrying struggles we cannot see. Today, I remember Anthony for who he was—not how he died: his kindness, humour, generosity, and the way he made people feel seen and valued. So please reach out to your colleagues if you think they’re struggling. You never truly know what someone is carrying when they show up to work each day.
-
Justin Norris liked thisJustin Norris liked thisCome work with me! ✨
-
Justin Norris reacted on thisJustin Norris reacted on thisLinkedIn allowing you to use AI to create content and to flag content as AI Slop
-
Justin Norris liked thisJustin Norris liked thisSometimes little moments of joy or whimsy come in unexpected places. Happy Friday! ☀️🍳
Experience & Education
-
360Learning
View Justin’s full experience
See their title, tenure and more.
or
Licenses & Certifications
Volunteer Experience
-
Toronto Marketo User Group Leader
Marketo
- 1 year 5 months
Science and Technology
I co-lead the Toronto Marketo User Group (TORMUG), bringing together Marketo users from around the GTA to talk best practices in marketing automation.
Honors & Awards
-
Marketo Champion (2018)
Marketo
-
Pipeline Marketing Award (2017)
Bizible
The Pipeline Marketing Awards were judged by a panel of marketing leaders from Uberflip, Radius, Linkedin, Heinz Marketing, and Bizible.
https://www.bizible.com/blog/2017-pipeline-marketing-awards -
Marketo Champion (2017)
Marketo
Champions are Marketo's most advanced customers who have demonstrated outstanding leadership in the Marketo Community and at Marketo events, are Marketo Certified Experts, are avid contributors in the social world, and are loyal advocates of the Marketo brand.
-
Bizible All Star (2017)
Bizible
The Bizible All-Stars are a select group of customer advocates who provide input into the future direction of our product.
-
Marketo Champion (2016)
Marketo
Champions are Marketo's most advanced customers who have demonstrated outstanding leadership in the Marketo Community and at Marketo events, are Marketo Certified Experts, are avid contributors in the social world, and are loyal advocates of the Marketo brand.
-
Marketo Champion (2015)
Marketo
Champions are Marketo's most advanced customers who have demonstrated outstanding leadership in the Marketo Community and at Marketo events, are Marketo Certified Experts, are avid contributors in the social world, and are loyal advocates of the Marketo brand.
Other similar profiles
Explore more posts
-
Traction Complete
Curious about some of the benefits around automating account hierarchies in #Salesforce? It's not just being able to reduce the manual effort needed to keep hierarchies up to date. Hear from Riley Smith, Lead Systems Architect at Klaviyo, about being able to leverage targeted enrichment and getting the most out of multiple data providers. Tune in below 👇 #revops
1 Comment -
CloudKettle, a Bell Canada company
Efficiency is a compounding interest. 📈 In his latest post, Greg Poirier explains why the Ateko Salesforce Practice is obsessing over a 5% quarterly gain—and how hiring for curiosity is our secret weapon for AI adoption. From "The Accountant Paradox" to the "Clean Sheet" advantage, see how we’re reshaping the way we work. Read it here: https://lnkd.in/eE3p4wRi
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