Luca Candela

I was CPO at Blend before their $4 billion IPO. We rejected the “forward deployed engineer” Palantir model that everyone is copying now. Here’s why: Blend could have followed the forward-deployed engineer approach. Both our CEO and CTO were early Palantir. They'd seen it work. Instead, we built a SaaS product with core components. Move fast, iterate, use data to learn. Not custom code for every client. That Palantir model is having a resurgence. Now Sierra, Decagon, and others are hiring forward-deployed engineers. But it’s not the right approach for every company. It’s fundamentally the wrong answer for Assembled’s customers. Here's what we understand about support teams and why we’re not going the forward deployed route: Customer Support teams are used to getting no resources and no support. On the surface, you'd think they'd be excited about engineers building custom solutions. "Finally, someone's offering us developers!" The reality is support teams are terrified of getting help they can't afford to keep. Support teams are wary because they know the math. They don't have budget for expensive, scarce resources. At some point the shoe drops. The engineers disappear when priorities shift. And custom solutions break. Support is left holding the bag with something they can't maintain. They're resource-starved by design. The idea of a dependency on resources they can't control is terrifying. So instead, the goal should be to give support teams the ability to manage, change, and solve their own problems. Not dependencies on engineers they can't keep. Smart, competent people who teach them to do it themselves. Palantir wasn't even a product company for years. They were a custom development shop with reusable components. Show up, build something bespoke, maybe extract learnings. That's not scalable for any but the largest support teams who need sustainability. Forward-deployed engineers are the new shiny object. I wasn’t a fan of it at Blend. And I’m not a fan today either. What I see in the customer support space is that teams need empowerment, not dependency. I want the teams we work with to be able to say: "I love that you’re partnering with me. And I love that I can do this myself too." That's the difference between building a product and building a consulting practice.

80

6 Comments

The hottest role in AI startups right now isn’t Forward Deployed Engineers. It isn't GTM Engineers. It’s Deployment Strategists. Decagon calls it an “Agent Product Manager.” Harvey calls it a “Solutions Architect.” Palantir Technologies has had versions of this role for years. And the salaries are climbing fast: - Decagon: $200k–$285k - Palantir Technologies: $120k–$200k - Figma: $150k–$260k - Ramp: $100k–$180k - Harvey: $190k–$260k So who are these people? They are usually pseudo-technical -- CS or engineering majors, or folks with technical work experience. Many come from 2 years in consulting, IB, or PE, then jump into startups to get their hands dirty. They are young, hungry, polished, and comfortable being in front of customers. What do they actually do? They make sure enterprise AI deployments succeed. A $100k+ deal does not survive on a nice pitch or a self-serve onboarding flow. It survives if the customer sees value in the pilot. That means: - Embedding directly with the customer - Designing prompt logic for specific workflows - Working with engineering to align integrations and data flow - Helping exec teams define their AI roadmap - Running feedback loops into product and GTM Why does this role matter so much? Because enterprise AI is messy. Integrations, data transfer, and adoption make or break a deal. Most buyers are using AI for the first time, and each has unique workflows. Deployment Strategists bridge that gap. They own the outcome. They are accountable for making pilots successful, which often means millions in revenue down the line. At Conversion, Sam Bochner has been leading this work for us. We are now thinking about scaling it into a full team. Because a few successful pilots can fund an entire department, and the cost of failed deployments is too high to ignore. Is this just a rebrand of customer success? Not really. Success is about answering tickets and renewals. Deployment Strategy is about going deep with a few enterprise accounts, extracting maximum value, and ensuring the pilot closes into a multi-year contract. Call it Agent PM, Solutions Architect, or Deployment Strategist. Whatever the title, this is becoming one of the most important roles in AI SaaS.

656

91 Comments

Columbus, Ohio has 13 percent growth in company formation YoY. We also just passed Chicago as the 5th largest concentration of data centers worldwide. Construction is accelerating. #ohiotech -

2

The simultaneous emergence of Claude Cowork and Clawdbot (now rebranded as Moltbot) is a clear signal of where consumer AI is headed this year: genuine end-user productivity through semi-autonomous agents. The days of purely prompt-based, one-shot results are giving way to systems that can handle real, ongoing work based on high-level requests. These tools are starting simple, but their trajectory is unmistakable: they will likely evolve quickly into capable digital assistants running 24/7. If recent AI trends hold, most of us will soon dedicate at least one (if not several) always-on machines or cloud instances to handling our routine personal and professional chores. This shift could free up significant time and mental energy, allowing us to focus on higher-leverage activities with far greater returns. Security, however, remains the single biggest hurdle. Granting AI agents access to local files, email, calendars, or messaging apps introduces real risks such as prompt injection, data leakage, unintended deletions, or worse. While tools like Cowork use sandboxes and careful permissions, and self-hosted options like Moltbot keep data local, vulnerabilities persist and demand vigilance. For now, the wisest path is cautious experimentation: start small, isolate sensitive data, monitor closely, and contribute feedback to help shape safer, more reliable agents. These early steps are opening exciting new chapters for everyday users. AI, it appears, won't just answer questions, but will truly work alongside us.

36

3 Comments

Alignment is the new limiting factor for AI-enabled product teams. Here are four lessons we have learned shipping with AI. Devs can code in days what would have taken weeks. If your internal comms don't keep pace, you just burn cycles and tokens with everyone moving at warp speed in different directions. Here are some hard-won lessons in how to keep things on the rails: 1. You need to guide the client’s vision. The pace of innovation is moving so fast it is a virtual certainty your clients don’t even know what they could be asking for. Your job is to help them get to the business outcome they need, which may mean thinking ahead of the ask. Chances are they are asking for half of what they really want because they have been trained to think the real ask is too big. 2. Good requirements are everything.  When you CAN build anything, you need to be laser focused on what you ARE building. Good requirements don’t mean immutable ones either. Sometimes requirements can change mid-project. This used to be a red-alarm, full-stop problem. Change can happen because of No.1 above, and you don’t ned to fear it because of No.3 below.     3. You can throw away bad code. Do it. You can now make working prototypes faster than you used to be able to make the first Figma of a plan. So do it. Write the code, publish the branch, test it. And if it fails? Throw it away. In the past, dev time was the most precious resource on a project. Now burning time imagining how the feature could work (instead of just making it) has the greater downside. 4. Most important: comm early, comm often. Teams need to be talking earlier and more often in the dev process. Misalignment about expectations, capabilities, velocity, breakthroughs and blockers can surface in hours versus days or weeks. The good news is, these can all be solved just as fast if you catch them. The bad news is, if you don’t catch them, they compound at an exponential rate. The big lesson here is that AI development will stress-test your internal and client-facing comms and processes like never before. AI can create organizational inefficiency as fast as it can create working code. If you can scale and refine your process to meet these challenges, though, insane velocity increases await. #enterpriseAI #AIdevelopment #vibecoding #AIforbusiness

3

Here’s a truth that often gets missed: becoming an AI-first company at the enterprise level is still really hard. It requires making big investments in R&D and specialized talent and betting on still-evolving frameworks and tools — and having the commitment to push the field forward yourself where the right tools or integrations don’t yet exist for your specific use case. As deploying AI and agents becomes table stakes in more industries, some companies end up hyping prototypes or easy use cases on earnings rather than building complex, self-improving AI systems to move the needle on productivity and profit. The task of separating these AI talkers from AI doers in public markets  – and leveraging AI to do it – is why my longtime partner John Hughes and I created the Swift AI Index as a fun side project last year. We just updated the index with five new stocks as part of a quarterly refresh; check it out: https://lnkd.in/eNBGjQzK

79

7 Comments

𝗧𝗵𝗲 𝗩𝗲𝗿𝗯𝗼𝘀𝗶𝘁𝘆 𝗣𝗿𝗼𝗯𝗹𝗲𝗺: 𝗪𝗵𝘆 𝗬𝗼𝘂𝗿 𝗔𝗜 𝗢𝘂𝘁𝗽𝘂𝘁𝘀 𝗧𝗼𝗼 𝗠𝘂𝗰𝗵 (𝗔𝗻𝗱 𝗛𝗼𝘄 𝘁𝗼 𝗙𝗶𝘅 𝗜𝘁) We've trained ourselves into a bad habit with LLMs: dump context, get a wall of text, copy what we need. But here's the irony—the verbose output that feels "thorough" often means the model is compensating for ambiguity by covering all it’s bases. 𝗧𝗵𝗲 𝗺𝗲𝗰𝗵𝗮𝗻𝗶𝗰𝘀 𝗺𝗮𝘁𝘁𝗲𝗿 LLMs don't "understand"—they predict the next token based on everything that came before. Tokenization fragments your intent into pieces the model reassembles probabilistically. Long, ambiguous prompts create more room for drift. Each completion builds on the last, compounding small misalignments into large ones. RLHF reasoning models feel smarter. They are. But even with emergent chain-of-thought processing, you may not see their reasoning—and when you do, it may not align with yours. The model optimized for "helpful" isn't the same as the model optimized for "correct given your specific goal." 𝗧𝗵𝗲 𝗳𝗶𝘅: 𝗣𝗿𝗼𝗴𝗿𝗲𝘀𝘀𝗶𝘃𝗲 𝗜𝗻𝘁𝗲𝗻𝘁 𝗗𝗲𝘁𝗲𝗿𝗺𝗶𝗻𝗮𝘁𝗶𝗼𝗻 Two instructions that change everything: 𝟭. "𝗕𝗲𝗳𝗼𝗿𝗲 𝗿𝗲𝘀𝗽𝗼𝗻𝗱𝗶𝗻𝗴, 𝗮𝘀𝗸 𝗰𝗹𝗮𝗿𝗶𝗳𝘆𝗶𝗻𝗴 𝗾𝘂𝗲𝘀𝘁𝗶𝗼𝗻𝘀." 𝟮. "𝗟𝗶𝘀𝘁 𝘆𝗼𝘂𝗿 𝗮𝘀𝘀𝘂𝗺𝗽𝘁𝗶𝗼𝗻𝘀 𝗯𝗲𝗳𝗼𝗿𝗲 𝗽𝗿𝗼𝗰𝗲𝗲𝗱𝗶𝗻𝗴." This surfaces the gap between what 𝘺𝘰𝘶 meant and what the 𝘮𝘰𝘥𝘦𝘭 inferred. Now you can align it with your actual objective before it generates 2,000 words in the wrong direction. Why this works: → 𝗥𝗲𝗱𝘂𝗰𝗲𝘀 𝗼𝘂𝘁𝗽𝘂𝘁 𝗯𝗹𝗼𝗮𝘁 — The model doesn't hedge when it understands your actual goal → 𝗔𝘃𝗼𝗶𝗱𝘀 𝘁𝗵𝗲 𝗨-𝘀𝗵𝗮𝗽𝗲𝗱 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 𝗽𝗿𝗼𝗯𝗹𝗲𝗺 — Critical details buried in long prompts get lost; shorter exchanges keep everything salient → 𝗜𝗺𝗽𝗿𝗼𝘃𝗲𝘀 𝘆𝗼𝘂𝗿 𝗼𝘄𝗻 𝘁𝗵𝗶𝗻𝗸𝗶𝗻𝗴 — Seeing the model's assumptions forces you to clarify your own 𝗧𝗵𝗲 𝗵𝗮𝗿𝗱 𝗽𝗮𝗿𝘁: 𝗬𝗼𝘂 𝗵𝗮𝘃𝗲 𝘁𝗼 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝗿𝗲𝗮𝗱 𝘁𝗵𝗲 𝗼𝘂𝘁𝗽𝘂𝘁. Human validation isn't optional. If you're skimming and accepting, you're not collaborating—you're gambling. And if the session has drifted too far? Don't try to correct mid-stream. Start fresh. A new conversation with clear intent beats a long thread full of compounding misalignment. 𝗪𝗵𝘆 𝘁𝗵𝗶𝘀 𝗺𝗮𝘁𝘁𝗲𝗿𝘀 𝗳𝗼𝗿 𝗮𝗴𝗲𝗻𝘁𝘀 When the model is just generating text, misalignment costs you time. When the model has tools—when it can send emails, modify files, update databases, provision resources—the intent gap has real consequences. An agent that misunderstands your goal doesn't just give you a bad answer. It takes bad actions. At scale. Automatically. This is why agentic systems require more than better prompts. They require intent verification, human-in-the-loop checkpoints, and architectures that make assumptions visible before tools execute. What assumptions is your model making that you've never seen? #AI #LLM #Agents #EnterpriseAI #PromptEngineering

5

Over the last few weeks, I’ve been running a series of live sessions for PMs, engineers, and AI-curious builders who want to go beyond buzzwords and actually reason about AI products. We focused on three themes: 🧩 1. Deconstructing RAG In this session, we broke RAG down into simple, practical pieces: - A clear retrieval + generation mental model for more reliable AI products  - Core building blocks: embeddings, chunking, vector stores, retrievers, and the end-to-end flow  - How to decide when RAG is the right answer – and when it isn’t  - The PM’s role in RAG quality: metrics, hallucinations, and experiments  - Real‑world constraints: latency, cost, data quality, context limits, and common pitfalls we see in the wild  🤖 2. Getting Started with AI Agents In this webinar, we covered: - What an AI agent really is, and how it differs from a traditional app or “just a chat UI”  - Agent architecture 101: tools, memory, reasoning loops, LLM orchestration - How to think about frameworks like LangChain - A step‑by‑step path to your first working agent: define use case → design → build → deploy  - Common pitfalls first‑time builders run into, and simple patterns that actually ship 🎯 3. Breaking into AI Product Management AI PM interviews today look very different from “generic PM” loops. - How AI PM interviews differ from traditional PM interviews  - The typical formats used by startups and larger tech companies  - The level of AI & GenAI understanding PMs are expected to demonstrate (without being ML researchers)  - How to structure answers across product sense, execution, and AI trade‑offs - Common mistakes candidates make – and what strong, nuanced answers look like We also ran live mock interview with participants and shared real‑time feedback on problem structuring, communication clarity, and AI/product depth. Watching these breakdowns gave people a “behind the scenes” view into how interviewers actually think. 🔭 What I’m seeing across all three Across RAG, agents, and AI PM interviews, a pattern keeps repeating: > People don’t just need more content on AI.  > They need better mental models and feedback on how they think about AI products. That’s the gap I’m trying to close through my work – combining AI literacy, product thinking, and career growth for PMs, aspiring PMs, and technical professionals.   --- 👥 What’s next – AI PM Cohort To go deeper into this, I’m also running an AI Product Management cohort for aspiring and current PMs who want structured practice, real product scenarios, and feedback. If you’re exploring a move into AI/GenAI product roles or want to strengthen your decision‑making as a PM, this cohort is designed to give you that combination of concepts, frameworks, and hands‑on application. Details on structure, schedule, and pricing are here: https://lnkd.in/gwMX-v-A

18

💥 I just found a foundational flaw in Claude. All models, all interfaces. Repro steps below: 1. Go to Claude settings 2. Attempt to update your email address 3. If you can’t find how to do it on your own, ask Claude how to change your Claude email address 4. Validate that Claude is not hallucinating by going to https://support.claude.com and searching for “How do I change the email address associated with my account?” 5. Shake your head and think “sorry users, I know we kind of slid past this when you signed up. Hope the email you used to sign up never gets compromised or needs to change – magic links or SSO are the only option (darn it). And if it does, hope you didn’t buy an annual PRO plan because no refunds / transfers even if to the same human ✌ (grrrr…)” Perhaps this behavior is the result of conscious prioritization based on a thorough understanding of needs / trade-offs. If this behavior was an oversight, then based on prior trips on the modernization train, this could be a massive slog of tech debt. Particularly in a hyper-growth business dependent on uninterrupted & high-integrity data access / entitlements / provenance. Product people – let's use this moment to build foresight! - Don’t abdicate PM or UX from “technical” discussions about data models, identity, or access control – decisions within these domains have incredible implications - Customer onboarding & admin workflows are foundational CX, not afterthoughts - If you’re using Claude Code, invest in Git + GitHub integration for improved resilience (and much more)

3

Software teams are rethinking the 2-week sprint. AI tooling has changed the equation: Engineers can ship features in hours or days while GTM still needs weeks to prepare launches. That speed asymmetry has fractured the shared rhythm that kept product teams aligned for the past decade. With AI writing code and teams deploying daily, the bottleneck has shifted. It's no longer "how fast can we build this?" It's "how do we coordinate releases when different parts of the org move at fundamentally different speeds?" Here are three approaches organizations are evaluating: 1. Continuous deployment and feature flags Who does this: High-performing tech companies (pioneered by MAANG 15+ years ago) Requires: Microservices, observability, automated rollbacks, flag management platforms Reality: About 75% of high-performing teams already use this approach. It's becoming table stakes. 2. Compressed cycles (1 week) Who does this: Small, autonomous teams with direct user access Requires: Minimal dependencies, tight scope, low-risk deployments Reality: This often increases meeting overhead without addressing the underlying coordination problem. 3. Extended cycles (6 weeks à la Shape Up) Who does this: Basecamp and select companies prioritizing deep focus Requires: Strong product shaping, clear boundaries, organizational discipline Reality: Limited adoption outside Basecamp. Trades Scrum ceremony overhead for another type of overhead. The insight most teams miss: The most effective approach (decoupling deployment from release) isn't new. It's been standard at high-performing teams for over a decade. But AI has made this practice necessary, not optional. When engineering can ship daily but your GTM motion requires weeks of prep, you can't keep deployment and release coupled without creating bottlenecks. The real challenge is organizational readiness. Most teams know best practices exist but they lack: → The architectural maturity to implement them → Feature flag management infrastructure → Clear ownership over release decisions → Strong product ops function → Coordination mechanisms that work across different speeds For most organizations, the answer isn't just adjusting sprint length. It requires investment in infrastructure, tooling, and organizational change. → I took a look at what's happening to the 2-week sprint and what it means for product teams. Read the deep dive: https://rebrand.ly/myf1p7u #ai #aicoding #productmanagement #aiproductmanagement #aiproduct #devops #productops #sprint

10

A product team we worked with was juggling multiple external vendors - development, design, QA - across different time zones. Deadlines were missed, communication was inconsistent, and the PMs spent more time chasing updates than building the product. We introduced an AI-powered vendor management tool that: - Monitors timelines and flags missed deadlines - Analyzes code quality and rework percentages - Tracks communication patterns, clarity, and collaboration - Measures response speed to ensure smooth coordination The outcome? - Managers had a single dashboard for all vendors - Problems were caught before they snowballed - Teams could focus on product delivery instead of firefighting The bigger picture: managing external teams isn’t just about keeping tabs, it’s about predicting issues and improving collaboration. At MintyMint, we use AI and data insights to help companies manage vendors seamlessly, reduce risks, and deliver faster. 👉 DM me if you’d like to explore it more. #MintyMint #AI #VendorManagement #Productivity #Collaboration #ProjectManagement #TechTools #DigitalTransformation

7

2 Comments

Execution capacity used to scale with headcount. Throw more people at the problem, get more done. That math has changed. A small team with strong AI orchestration skills will outpace a large team without them. Every time. The new constraint is finding the right people. That's the bottleneck now.

18

4 Comments

End users ain't prompt engineers. Appropriate prompts play a crucial role in guiding Language Model Models (LLMs) towards accurate outcomes. The effectiveness of LLMs often hinges on precise prompts, continuous refinement, and diligent result evaluation to ensure the model stays on course. However, in practical scenarios, users exhibit higher expectations and may not possess the capacity or willingness to actively collaborate with the LLM. This challenge is perhaps most evident in customer service interactions, where end users can be impatient and frequently disgruntled. Real-world bots encounter a dual challenge. Firstly, they must navigate through vague initial prompts, meaning the bot has to iteratively refine and clarify user intentions, thereby increasing the interaction complexity. Secondly, maintaining the correct context across multiple interaction stages poses a significant hurdle, as highlighted in this work by Microsoft and Salesforce. Model drift, stemming from inaccurate initial assumptions, can escalate over time, particularly with intricate requests. Additionally, prolonged interactions may amplify user frustration, leading to less clear prompts—an extremely challenging scenario for bot performance. Addressing these challenges is poised to be a key research focus in the upcoming year. Exploring how end users can enhance their prompt engineering skills or better still, developing LLMs capable of adeptly handling complex, ambiguous, multi-step requests without extensive user input are crucial objectives for future advancements in this field. For more insights, refer to the research paper: https://lnkd.in/gkTT5bkz #AI #promptengineering #chatbots

43

6 Comments

GTM goals for early stage technical companies: 1. Be referenced often in internal meetings of your prospects (exploratory phase of adoption) 2. Be referenced often in externally facing architecture diagrams of the top companies in your market (first path to production + PMF) 3. Be endorsed often in communities as experts/influencers see 1 + 2 appear without your promotion (your real ARR + NRR story happens here) 4. Be endorsed as the gold standard by communities (moat, hang on for dear life and keep shipping and making money) #2 is v hard as things shift often, but Founder time should be with customers to make this occur as often as possible. #3 + #4 become your roadshow focus over time so you can scale your company.

21

2 Comments

For most of the last twenty years, one of the smartest things you could build was a software company. But something is shifting. The barrier to building software is collapsing, and the real leverage may be moving to the people who know how to use these tools to fix real problems inside businesses. I just published my first ever Medium article why the SaaS dream that shaped the last decade may be fading, and why the next era might belong to the people who just figure things out. Would love your thoughts. https://lnkd.in/efSttmvP

43

6 Comments

I wrote about how product managers should deliberately decide which features should be built to an A grade and which ones should be built to a B or even only a C grade. Making the resource tradeoffs to optimize business outcomes is a core job of product management, and building every capability to an A grade is a poor use of scarce resources. https://lnkd.in/gTXBfNYE

31

4 Comments

AI is creating a new execution risk: two communication protocols AI is not being “adopted.” It is being embedded into the workforce. That turns every company into a hybrid org, humans + agents, and creates a new execution risk: everyone becomes a manager. Design for it and you scale execution. Ignore it and you scale rework + cultural erosion. Hybrid orgs need a hybrid operating model. Two challenges show up immediately: 1) Management becomes a baseline skill Every employee will need to assign work, set standards, review output, and improve performance because agents become integral to deliverables. 2) Communication splits into two lanes Agents need precision: context, constraints, acceptance criteria, escalation paths, measurable feedback loops. Humans need trust, nuance, psychological safety, and space for disagreement and learning. The analogy that keeps coming to mind: global scaling. When teams span cultures and time zones, one communication style stops working for everyone. You have to be intentional about how you communicate, what belongs in writing vs live conversation, what needs specificity vs nuance, how you avoid misread tone, and how you keep collaboration strong when you can’t rely on real-time back-and-forth. Hybrid human + agent teams create the same requirement except the “other culture” is machine logic. If you don’t clearly separate the modes, people start pushing agent-style communication onto humans (or expecting human-like intuition from agents). Both are failure modes. I loved how Bethany Ayers discussed this on The Operations Room podcast (Episode 91, AI in the Future of Work): agent management becomes a core capability for every employee, and people will increasingly be measured by the outcomes their agents produce. In some orgs, AI adoption is already showing up in promotion and incentive conversations. I have built operating models across startups and global orgs, this is the same pattern: when the communication protocol changes, you have to design the norms or you pay later. Awareness is step one. Step two is building muscle: what “good agent management” looks like by role, how employees are trained, and how leaders reinforce the two-lane communication model. This is the work I do as an operator: designing operating models that scale without cultural drag. If you are a CEO/COO/CHRO navigating this shift (especially in a scaling org), I am happy to compare notes and share a simple way to operationalize it without adding bureaucracy. #COO #Operations #OrgDesign #FutureOfWork #AI #Leadership #Culture

31

3 Comments