About
Outgrowing your platform?
The rewrite conversation is usually a symptom, not the…
The rewrite conversation is usually a symptom, not the…
Activity
2K followers
-
Brian Wasserman posted thisOne developer vibe-coding is a bug. Fifteen developers vibe-coding is a platform you can't maintain. Early in my career I watched several product teams independently build their own checkout flow — each one integrating separately with the same internal payment system. Seven integrations. One payment library. Every team had made a reasonable decision in isolation. Nobody had looked left or right. The cost showed up later. Every time we added a payment type, all seven teams had to make the same change, retest, and re-release independently. A processing change meant seven parallel efforts. The checkout, cart, and confirmation pages weren't consistent across products — functionality differed, the experience differed, and we left value on the table because of it. When we finally consolidated, it took significant time and resources that could have gone somewhere else. The ungoverned complexity had been accumulating quietly the whole time. That was before AI could generate thousands of lines of code in minutes. Ungoverned complexity doesn't just accumulate anymore. It compounds exponentially. I ran an experiment recently. Same problem, same AI, two approaches. Ungoverned — I gave it a prompt and got out of the way. It produced 37 files, invented features I never asked for, and optimized for architectural completeness. Governed — I constrained every step, defined roles, made explicit calls about what to build now and what to defer. Six files. Exactly what was needed. One developer doing this ungoverned is a bug. It'll surface eventually and someone will clean it up. Fifteen developers doing this ungoverned is a platform you can't maintain. The complexity accumulates faster than anyone can see it, and by the time it shows up in slowing velocity or mounting bugs, it's already deeply embedded in the codebase. AI doesn't degrade without oversight. It expands. And it will expand consistently across every developer on your team who isn't constrained. Governance isn't a process tax. It's the thing that keeps fifteen individual reasonable decisions from becoming an architectural problem nobody can explain. In the age of agentic AI coding teams, that's the difference between a platform that scales and one that quietly collapses under its own complexity.
-
Brian Wasserman posted thisWe paused a major rollout mid-stream. Every week we were stopped cost us real revenue. It was still the right call. I knew that because I wasn't just in my own lane. We had just released a major architectural change. First few weeks looked clean. Then reports started bubbling up from customers. Small things at first. Edge cases. But they pointed at the architectural change we'd just shipped. When certain people reached out, I knew to pay attention. I called the head of support and the head of customer success to understand what they were hearing from customers. The team investigated, reviewed the logs, and looked for the pattern. What they found was an issue happening more frequently than we realized — one that would continue to drive up support call volume and require internal resources to continuously clean up. We had a flag in the system that let us revert customers to the old architecture. We used it. Then we halted rollouts entirely. This mattered because we didn't get paid until a site went live. Every week of paused rollouts had a direct revenue cost, and the GM knew it. But runaway support volume had a cost too. Customer churn had a cost. And for our larger accounts, the downstream consequences of getting this wrong were significant. When the head of product and I walked into the GM's office together with a recommendation to pause, he understood the seriousness. We spent four weeks fixing, re-releasing, rolling out cautiously to a handful of locations, monitoring, then moving forward. The CTO who stays in the technical lane never gets to make that call. They're not in the room when support surfaces the pattern. They're not the person the support leader calls directly when something feels off. But it goes beyond incident response. The best technical decisions I made weren't made in architecture reviews. They were made in conversations with the support team about what was actually breaking for customers. In product discussions where business priorities had to get weighed against what the platform could actually sustain. In strategy sessions where revenue pressure was real and someone had to translate what "pause the rollout" meant for the business — and why it was the right call anyway. Dashboards tell you if the system is up. They don't tell you what's quietly eroding customer trust, which tech debt is starting to show up in support volume, or which architectural decision is about to become a churn problem. Dashboards drive reactive behavior. The intelligence that lets you get ahead of problems lives in other lanes. You have to be in those conversations — understanding what's driving the business, what's frustrating customers, what sales is promising and what support is absorbing — so that when you sit down to plan the engineering roadmap, you're building toward something real. If you're not in those lanes, your roadmap is just a list of things your team wants to build, detached from what the business actually needs.
-
Brian Wasserman posted thisThe math on offshore engineering made sense before agentic AI. Context just became the most expensive thing in your stack. We once handed a greenfield product to an offshore team. Good engineers. Clear ask. Six months to a prototype. They came back with a product direction that wasn't what US leadership had envisioned. More late nights and early mornings trying to realign. When they proposed an architecture, it leaned on shared code from an existing product — exactly the coupling we'd said was a non-starter. We had to stop. Pull in a US-based product manager and architect. Rebuild the context from scratch — the product's purpose, the MVP, the business goals, the architectural constraints. We spent more time defining context for the offshore team than we ever would have with a team sitting in the same building. It wasn't their fault. The product was solving a problem for a market in another continent they were not familiar with, and context doesn't travel well across a twelve-hour timezone gap Offshoring made sense when coding was the currency. Agentic AI made coding a commodity, and shifted the costs to context. Agentic AI teams run on context. The prompts, the governance rules, the architectural constraints — these are only as good as the context behind them. A senior engineer sitting close to the product, the customers, and the business goals can build and lead an agentic team for a fraction of what an offshore team costs. And they bring the one thing you can't delegate: context. The offshoring model was always a context problem. AI just made it impossible to ignore.
-
Brian Wasserman posted thisThe question that got me hired was about building features. My answer was about not building them. It was my first real executive interview. They asked how I thought about product decisions. I said: knowing what not to build is just as important as knowing what to build. Something shifted in the room. I meant it. And I had a story to back it up. Early in my time leading engineering, product came to me with a request for real-time reporting. Data flowing through the system, reflected in dashboards instantly. Technically possible. The team was excited about it. I killed it. At least for phase one. Real-time reporting isn't just a feature — it's a different architecture. Streaming infrastructure, higher complexity, a significantly longer delivery timeline, and AWS costs baked permanently into every customer contract. And nobody had done the ROI work. We were being asked to build expensive infrastructure without knowing whether customers would actually pay more for it. That's not a product decision. That's a bet. Instead I pushed for near-real-time. Reports updated within 10-15 minutes. We added a single line to the dashboard: "Data does not reflect transactions within the last 15 minutes." Customers were delighted. Nobody asked for real-time after that. The label set the expectation and the expectation was met. Months of engineering time saved. Architecture kept clean. Problem solved. But the harder version of this conversation happens when the first enterprise customer shows up. Up until that moment the team has been building a product — features that serve everyone, decisions made with the whole user base in mind. Then a big customer arrives with a list of requests and a large contract attached. The temptation is to build whatever they want. That's where a lot of companies quietly lose the plot. They start building for one customer instead of the product. Features that won't generalize. Workflows that only make sense for one org chart. Technical debt that arrives dressed as revenue. The right question isn't "can we build this?" The answer to that is almost always yes. The right question is "should we — and for whom?" Saying no to the wrong features is how you protect the yes that actually matters.
-
Brian Wasserman posted thisA college sophomore stopped me at a networking event last week. He's building his first product. Computer science major, one-man shop, using AI to move fast. Excited about what he can build. We talked about his architecture. I gently asked if he was reviewing the choices AI was making — not just accepting the output, but actually understanding whether those choices were right for where his company is today. He thought about it. Then asked me a question I wasn't expecting. "How do I learn to think about this the way you do?" I didn't have a comfortable answer. You don't learn it from a course. You don't learn it from a book. You learn it by building something that breaks under pressure, by making an architectural decision that felt right at the time and living with the consequences two years later. By being the person in the room when the rewrite conversation starts. I told him: this is what battle scars are. AI can generate the code. It can propose the architecture. It can move faster than any team I ever managed. What it can't do is tell you whether the architecture is too complex for a two-person startup, or too rigid for the pivot that's coming, or creating a coupling problem that won't show up for eighteen months. That judgment comes from having been wrong before.
-
Brian Wasserman posted thisI spent two days building an AI engineering team. I wrote zero lines of application code. A product manager agent wrote the functional spec. An architect agent designed the system. A senior engineer agent created the implementation plan. A junior engineer agent wrote the code. A tester ran it. A debugger fixed the failures. My idea prompt was six lines. The team did the rest. Here's what surprised me. I didn't spend my time on the product. I spent it on governance — defining each agent's role, their constraints, what they were and weren't allowed to do. At one point the debugger decided to rewrite the test suite instead of the implementation code. It wasn't wrong. But it was a judgment call a human developer and tester would have made together. Without governance rules defining that preference, the agent took a shortcut that could have been a problem. That's the moment that changed how I think about this. A non-technical founder running this same experiment would have accepted that output without knowing what question to ask. Is the architecture too complex for where the company is today? Does the code create a seam problem six months from now? Is the test rewrite masking a deeper issue? AI doesn't answer those questions. It executes within whatever boundaries you give it. I didn't stop being an engineer. I just stopped writing code and started writing the rules the code has to follow.
-
Brian Wasserman posted thisThe question that got me hired was about building features. My answer was about not building them. It was my first real executive interview. They asked how I thought about product decisions. I said: knowing what not to build is just as important as knowing what to build. Something shifted in the room. I meant it. And I had a story to back it up. Early in my time leading engineering, product came to me with a request for real-time reporting. Data flowing through the system, reflected in dashboards instantly. Technically possible. The team was excited about it. I killed it. At least for phase one. Real-time reporting isn't just a feature — it's a different architecture. Streaming infrastructure, higher complexity, a significantly longer delivery timeline, and AWS costs baked permanently into every customer contract. And nobody had done the ROI work. We were being asked to build expensive infrastructure without knowing whether customers would actually pay more for it. That's not a product decision. That's a bet. Instead I pushed for near-real-time. Reports updated within 10-15 minutes. We added a single line to the dashboard: "Data does not reflect transactions within the last 15 minutes." Customers were delighted. Nobody asked for real-time after that. The label set the expectation and the expectation was met. Months of engineering time saved. Architecture kept clean. Problem solved. But the harder version of this conversation happens when the first enterprise customer shows up. Up until that moment the team has been building a product — features that serve everyone, decisions made with the whole user base in mind. Then a big customer arrives with a list of requests and a large contract attached. The temptation is to build whatever they want. That's where a lot of companies quietly lose the plot. They start building for one customer instead of the product. Features that won't generalize. Workflows that only make sense for one org chart. Technical debt that arrives dressed as revenue. The right question isn't "can we build this?" The answer to that is almost always yes. The right question is "should we — and for whom?" Saying no to the wrong features is how you protect the yes that actually matters.
-
Brian Wasserman posted thisOne engineer could take down the entire platform. Not through malice. Just by deploying a routine change. I once joined a team that had built something real. Shipped fast, landed customers, grown the product into something people depended on. The architecture reflected every decision that got them there. One core service. Everything in the system ran through it. Every process, every operation, every piece of data read directly from its internal memory structures. When it worked, it was elegant in its own way. When it didn't, it was a five-alarm fire. At scale it crashed constantly. With everything running through a single service, high load turned into a debugging nightmare — dozens of async operations in flight, no clean way to isolate what caused the failure. Fix the crash, redeploy, hope. The coupling made it fragile in a different way too. Change one part of the system and something completely unrelated would break. Fix one bug, introduce another. The team wasn't slow because they were bad engineers. They were slow because the system punished every move they made. And scaling it was eye-watering. No horizontal scaling — the architecture couldn't support it. Just vertical scaling across multiple deployment stacks. The AWS bill reflected every shortcut taken three years earlier. The painful part wasn't the re-architecture. It was knowing how much of it was avoidable. Basic architectural seams early on wouldn't have slowed the team down much. A little separation between components. Some unit tests. Clean enough code that the system could be understood by someone who didn't build it. Not perfection. Just enough structure that the system didn't fight you when it was time to grow. Moving fast is the right call in the early days. But fast and fragile aren't the same thing — and the difference only shows up later, when it's expensive.
-
Brian Wasserman posted thisI once spent three and a half weeks making a two-day code change. Not because the problem was hard. Because nobody had written down how to build the component. About 15 years ago I was asked to fix a bug in the payment processor. I was a Java engineer. They pointed me at the component and said go fix it. That's when I discovered the problem. The source code lived in a folder on someone's local machine. Never checked into source control. The build instructions existed nowhere — not in a wiki, not in a doc, not in an email thread. The last engineer to build this component had done it almost a year earlier, gotten a new PC since then, and couldn't even compile it themselves anymore. Before I could change a single line of code I spent a week reconstructing how to build the component from scratch. Then reverse engineered its functional requirements because the only spec was the code itself. Then got everything documented, checked into source control, and deployed just to confirm my version behaved the same way as what was already live. All of that before I could even make the fix. Three and a half weeks. For a two-day fix. And the whole time I kept thinking about the same thing: What if this hadn't been a planned change? What if the payment processor had just crashed? Production down. No one can register. No one can pay. Support is flooded. The CEO is on the phone with angry customers. And somewhere an engineer is doing exactly what I did — chasing people down hallways, reconstructing institutional knowledge from scratch, hoping the one person who knows something still works there. That's a five-alarm fire with no fire extinguisher. This doesn't show up on any dashboard. Nobody files a ticket for "critical knowledge living only in Dave's head." But it's real. And at some point Dave leaves. Or Dave gets a new laptop. Or the system crashes on a Saturday and Dave is unreachable. Every startup accumulates this. It's almost unavoidable when survival is the priority and moving fast is the only mode you know. The question isn't whether it exists in your company. It's whether you know where it lives before you need it.
Brian Wasserman commented on a post
17h
Congratulations Amanda!
Brian Wasserman replied to a comment
2d
Matt Haney Its more expensive unfortunately to design this in later.
Brian Wasserman commented on a post
2d
This is the exact problem I've been thinking about with AI engineering teams too.
When you give an AI agent a task with no pushback mechanism, it just executes. Every idea, every direction change, no resistance. Sounds great until six months later you have 37 files nobody understands.
I built an Architecture Council into my agentic pipeline for exactly this reason. Three agents whose only job is to object before a single line of code gets written. They push back on the architect. They require changes. They block bad decisions.
It's the same job a good engineering leader does with a founder who has ten ideas before breakfast.
The seat that holds the line has to be designed in — whether your team is human or AI.
-
Brian Wasserman liked thisBrian Wasserman liked thisHow is Salesforce’s Agentforce impacting the government? For years, interacting with the government has felt slow, fragmented, and impersonal, not because people don’t care, but because the systems haven’t kept up. That’s starting to change. In my latest #sponsored episode of CXO Spice, I reconnected with Paul Tatum, EVP of Global Public Sector Solutions at Salesforce, to explore how agentic AI and Agentforce are moving from concept to reality across governments worldwide. A few things stood out: ✅ 82% of government organizations have already adopted AI agents (IDC, 2026) ✅ 56% believe agentic AI will have a more profound impact than the rise of the internet (IDC, 2026) ✅ Missionforce is turning agencies like USDOL, VHA, and the U.S. Army into agentic enterprises ✅ Trust, security, and compliance are built into the foundation And the real shift is experiential. We’re moving from reactive systems to proactive, personalized government services that anticipate needs and reduce friction in everyday moments. If the past decade of government was defined by digitization, the next will be defined by intelligence. Missionforce and Agentforce are the catalyst, and the real impact is a future where the government anticipates, adapts, and serves at the speed of its citizens. And for the first time in a long time, the gap between what the government is and what it could be for the citizens is finally closing. That is Agentforce at work for the government. Read the full newsletter + watch the episode: https://lnkd.in/gd-fTkFY #AI #GovAI #Agentforce #Missionforce #SalesforcePartner To stay current with the latest trends in #Technology and #Innovation, Subscribe to 👉 #CXOSpiceNewsletter here https://lnkd.in/gy2RJ9xg or 👉 #CXOSpiceYouTube here https://lnkd.in/gnMc-VpjHow is Salesforce Empowering the Government with Agentforce?How is Salesforce Empowering the Government with Agentforce?
-
Brian Wasserman reacted on thisBrian Wasserman reacted on thisNo commute No office politics No begging for annual leave No 250 applicants chasing the same role No six-month probation to prove yourself — again No culture fit interviews with people half your experience No lying awake wondering if your name's on the next restructure list This is what fractional work looks like for senior leaders Directors. VPs. C-suite executives with ten or more years of proven delivery Working on £5–10K retainers per month Three or four clients Total earnings of £180–300K — on their own terms Here's why smart companies are quietly moving this way 50% of full-time hires fail within 18 months That's not a recruitment problem That's a structural one Full-time hiring is slow, expensive, and risky — but the need for specific expertise on specific projects doesn't disappear So instead of a permanent hire, they bring in fractional experts People who've already led this work — in finance, marketing, sales, operations — at other organisations People who arrive ready No ramp-up No hand-holding The problem isn't the model It's the mindset Most senior leaders never consider this because they've been quietly conditioned to believe that employment is the only legitimate way to monetise a career One employer One income stream One identity wrapped around a job title And that's the trap Right now, you have one client -Your employer They control your income, your schedule, your professional sense of self And if they decide to restructure — your income goes to zero overnight The shift is simple, even if it doesn't feel that way Stop thinking of yourself as an employee Start thinking of yourself as a business — and your expertise as the service You don't need a new idea You don't need a startup pitch You need to capitalise on what you've already spent a decade building Three or four fractional clients The work you already know how to do Something that finally belongs to you That's the model And for the right people — it's not just viable It's transformational, Has for me..!
-
Brian Wasserman liked thisBrian Wasserman liked thisThere's a conversation most engineering leaders have lost at least once in the last 3 years. You know the foundations need work, but stakeholders want a new shiny feature (probably AI-related). You make the technical case, but you lose either way. Priscilla Nagashima, VP of Data and AI at Pleo, calls what happens next a “graveyard of apps.” Features that break in six months because the data was never reliable to begin with. I asked her how she wins that argument, her tip is to stop making it a technical debate and reframing it as a risk instead: "We can build the flashy thing now, and it'll break in six months. Or we can spend three months on foundations and build something that actually scales." Risk lands differently in a boardroom than best practices do. In my full interview with Priscilla, we also talked about: 🟡 Why most AI projects fail before the model is ever the problem 🟡 Why "culture fit" is quietly degrading your hiring decisions 🟡 How to evaluate remote and global candidates without relying on gut feel 🟡 The one skill that's become fundamental in the age of AI Read the full interview and subscribe for weekly insights from seasoned engineering leaders (link in comments👇)
-
Brian Wasserman liked thisWhat a great honor to join you and share the Olo (first 21 years) story, Rick! Thank you for sharing with your audience. I love this industry and its role in providing hospitality in a world that desperately needs more of it! If Team Olo and I can continue to be a force multiplier for hospitality, that’s a noble use of our time and talent. Hospitality is worth fighting for!Brian Wasserman liked thisFounders Journey -- A Conversation with Noah GlassThe Founder's Journey -- A Conversation with Noah GlassThe Founder's Journey -- A Conversation with Noah Glass
-
Brian Wasserman liked thisBrian Wasserman liked thisYes, it's the busiest time of the year. Yes, there's also dinner catering from CaterCow. Push hard for the next two weeks, we'll make sure every minute counts with on-time, delicious (and nutritious) food catered to the office.
-
Brian Wasserman liked thisBrian Wasserman liked thisPeople ask what fractional CTO work actually looks like. Here's Jon's Tuesday last week. 9am: Call with a CEO. Their VP Engineering just handed in notice. Two weeks to figure out interim leadership and start a search. Talked through options, calmed nerves, made a plan. 10:30am: Architecture review for a client preparing for Series B. Found a scaling bottleneck that would've cost them six months and £600K if it surfaced during due diligence. Now it's a fixable problem, not a fundraising crisis. 1pm: Sat in on an engineering leadership meeting. Didn't say much. Watched the dynamics. Spotted why decisions keep getting stuck. Gave feedback to the CTO afterwards. 3pm: Wrote a board update for a client whose CTO struggles with investor communication. Four slides. Clear narrative. Business language. 4:30pm: Call with a CFO about cloud costs. Helped them ask the right questions. Identified £180K of annual waste in 45 minutes. No code written. No Jira tickets closed. But five companies in better shape than they were yesterday. That's fractional CTO work. Strategic air cover for people who need experience on tap, not on payroll. What would change if you had that kind of support on a random Tuesday?
-
Brian Wasserman liked thisBrian Wasserman liked thisHighlight from our Monthly Collective Gathering SDLC → ADLC (Agentic Delivery Life Cycle) AI isn’t just accelerating the SDLC — it’s forcing a redesign of it. What’s emerging is an ADLC, where the constraint is no longer model capability, but system design: context pipelines, data accessibility, guardrails, and ultimately trust. The organizations pulling ahead aren’t those with the best models — they’re the ones structuring their data and workflows to make agents actually effective. Thanks to Guy Eisenkot (Baz) & Anish Agarwal (Traversal) for leading the conversations. Grateful to the leaders from: eBay, Harness, John Deere, Shopify, ING, Nomura, Fidelity Investments, Webflow, JPMorganChase, Upwork, and others.
-
Brian Wasserman reacted on thisBrian Wasserman reacted on thisOn June 5, 1978, at 19 years old, I started my career in technology as a computer operator at NCR Corporation. Within 18 months, I was offered the opportunity to travel to Cairo, Egypt on a 6-month assignment to deploy a major IT system at what was then the largest bank in Africa. I said “No.” I had never been on a plane and had barely traveled more than 100 miles from my home in San Diego. But a colleague stopped by not long after and told me how lucky I was. She had studied Egyptian culture and asked if I would bring pictures back. That conversation stayed with me. Within a day, I went back to my manager and said, “Sure, I’ll do it.” During that experience, I discovered I loved working across cultures and around the world. And the words “Sure, I’ll do It” became a theme of my career—from NCR, through roles at Dell and Intel, and ultimately to NVIDIA. For that first decade, I spent a lot of time wondering if I truly belonged in an industry surrounded by such smart people. By my early 30s, I came to an honest realization: I wasn’t going to be a great software engineer—and that was okay. What I did discover was that I had a passion for leading teams. I’ve always been drawn to people I considered very intelligent. Not because I considered them to be a peer group, but because I genuinely enjoyed learning from them. The conversations are better, the ideas are bigger, and the outcomes are more meaningful. I also learned that talented people want to work on difficult problems. So I made a habit of raising my hand for the most complex projects. Those choices allowed me to recruit incredible engineers and spend my career working alongside people who continually inspired me. Many years ago, someone said to me: “No one looks around at their retirement party and wishes they had done one more project. They think about the people they worked with.” He was exactly right. For me, it was about the people—the ones you built things with, learned from, supported, and grew alongside. And I was lucky to be a leader. It wasn’t’ complicated. Be kind. Treat people with respect. Be honest. At my last birthday, it hit me that I’m closer to 80 than I am to 50. That made this decision feel clear—this is the right time to step into the next chapter. So here I am, nearly 48 years later, retiring today and reflecting on a career I could not have imagined when I started. If I’ve learned anything, it’s this: • Say yes to opportunities—even the ones that scare you • Seek out meaningful challenges—they lead you to being part of incredible teams • The best leaders build trust, lift others up, and create environments where people can do their best work I’m looking forward to what’s ahead. And I’m trusting all of you who tell me that in retirement you are so busy you don’t have time to work. To everyone I’ve had the privilege to work with across these years and companies—thank you. I’m deeply grateful to have been a part of this amazing community…
Experience & Education
-
Wasserman Systems
View Brian’s full experience
See their title, tenure and more.
or
Volunteer Experience
-
Karate Instructor
Family Karate, Rancho Penasquitos
- Present 13 years 4 months
Children
Assisting / Teaching Karate to children and adults with special needs.
-
Coach
Shooting Stars Basketball
- Present 2 years 9 months
Children
Volunteer coach for special needs basketball
Recommendations received
18 people have recommended Brian
Join now to viewOther similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content