Bobby Beckmann

🚨 BREAKING: Confluent is looking to be acquired, sources say. Confluent is allegedly working with an investment bank on the process of being sold "after attracting acquisition interest". Reuters broke the story, citing three people familiar with the matter. Are you surprised? I'm not. Confluent’s future wasn't looking bright: • increased competition (Kafka is becoming commoditized) • big customers reducing spend (as shared in last earnings) • open sourcing of Diskless & Iceberg Topics (closing the latest proprietary feature gaps) • peanuts revenue from Flink (just $10M and recent) • stock performance has been an absolute failure (objectively speaking) A few months ago, its stock price cratered -40% overnight. As I mentioned back then, Confluent needs a miracle. A miracle in Flink/DSP spend to propel the company back into the rocketship it was. The numbers so far haven't shown that. 😔 Also, given the history of such attempts and clear lack of revenue to show for it - it wasn't looking pretty. The company marketing also gave it away. Marketing narratives included things like "beating anybody on price" (despite being famous for costing 2-3x more) or "building meal prep AI agents using 6 Flink jobs" 😂 Rumors about potential acquisition had been going on for a while, but they were just that - rumors from people who think that was the only way out. There were signs. 👉 Six months ago, Confluent overhauled its executive severance policy to favor "change of control" circumstances. That gives out full equity vesting & good payouts if execs get fired post-acquisition. As a reminder - no deal is guaranteed. There were similar rumors of a Snowflake 🤝 RedPanda acquisition half a year ago and that went nowhere. In any case, major consolidation in the streaming market is inevitable in the next 5 years. 🔮 You can quote me that I expect all major Kafka players you know today to be gobbled up by someone. I again repeat what I said a few months ago - acquisition would be a sad way to end Apache Kafka's commercial journey. Kafka benefits from a strong Confluent and Confluent benefits from a strong Kafka. Confluent is THE reason Kafka is where it is today. But that's how life turns out sometimes. 💔 Confluent could remain independent and it'd continue to live well. It has ~$900m in cash and is non-GAAP profitable. It can last a long time. But will it grow and perform well? 🤷‍♂️ My bet, though, is it'll get sold. Who do you think is buying $CFLT? #Confluent #ApacheKafka #CFLT #EventStreaming

408

65 Comments

Adding Document Understanding to Claude Code 📑🤖 Coding agents are awesome but they don’t natively understand docs (they kind of do, but it sucks). This limits their ability to build business applications - think about anything from a financial copilot to contract reviewer to rfp synthesizer to synthesizing PRDs. I wrote a blog to capture core patterns for adding document understanding to Claude Code - applicable to coding agents in general. This both helps Claude Code 1) get enterprise context when building apps, and 2) allows it to build more agentic, robust apps in general. 1️⃣ Add an MCP server that preindexes your file knowledge base 2️⃣ Let Claude Code loose on your filesystem with the right CLI tools for understanding files. 3️⃣ Give Claude Code context on core modules for document understanding and agentic workflows - this last step is important if you want to build “better apps in general” and not just “rule-based apps”. Blog: https://lnkd.in/g4EEFftk The blog offers general principles regardless of whether you’re using our stuff, because we fundamentally think that coding agents + docs leads to a ton of useful applications. Of course we are deeply investing in tooling in this space too (some referenced in the blog), a lot of it dependent on LlamaCloud. Check it out: https://lnkd.in/g9Wpqn7w

210

3 Comments

Validating Risk Scores at the Source Our HCC-In-FHIR package keeps getting more robust every day. Though I'll admit - the name's a bit misleading now since we've expanded beyond just FHIR (see the figure below). The package now handles three primary formats: X12 837, FHIR, and custom formats. But here's the key: it calculates CMS HCC risk scores directly from your source data. Why does this matter? Every data transformation step between your source data and final risk scores creates two problems: - Delays in getting actionable insights - Information loss along the way Anyone who's tracked a single claim from practice to CMS encounter data knows exactly what I'm talking about. The amount of data that gets lost from source to destination is staggering. By calculating risk scores directly from the data you receive from clearinghouses and submit to CMS, you get true visibility into what's actually happening - and what the real delta is. The package remains open source (https://lnkd.in/ebEjSqAc), but if you need help implementing it, we're here. At falcon health, we're already helping many MAPDs gain full visibility into their claim journey and HCC performance (feat. Prince Baawuah, DrPH, MPH, CSM®) Interested in learning more? Let's connect!

42

5 Comments

The EnFi, Inc team just launched another major customer! I’m still amazed seeing in action what production-level Agentic Architecture unlocks for both customers and platforms. What it means in practice: No more months of ETL. Connect, understand, act, day one. Immediate value: Immediate ~30%+ time and efficiency compression through automation across the process. Step-function jump: Application processing from 1–2 days ➜ ~7–10 minutes. In software terms, that’s wild. Flexes to new use cases. We’re no longer stuck in the rinse-and-repeat of onboarding ➜ product ➜ engineering to ship value. Feedback is built-in. Agents and users surface issues in flow; improvements feed straight into the pipeline. Fixes in days or weeks, not months or release cycles. That’s PLG, the way it should be. Super excited for what’s next, and incredibly grateful to our earliest customers for seeing (and shaping) what’s possible. We are live, this is real, and we are scaling a digital workforce for credit. And it makes me realize, Agentic is the new paradigm for software and product led growth. #AI #Agentic #Fintech #Lending #PLG #Automation #TimeToValue

76

3 Comments

AI in IT Ops is splitting into two camps - and your strategy decides your outcomes. 1. Reactive / Assistive AI Adds intelligence inside existing workflows: ticket triage, summarization, alert deduping, faster RCA. It accelerates humans and trims MTTD/MTTR - but only after something breaks or a user raises a hand. 2. Proactive / Autonomous AI Continuously watches telemetry, spots weak signals, predicts incidents, auto-remediates drift, tunes capacity before users notice. It reduces tickets altogether, not just handles them faster. Why it matters: - Fewer outages > Faster fixes - Prevented tickets free cycles for strategic work - Continuous optimization lowers infra & licensing waste - Better employee experience (issues “never happen”) Question for IT leaders: What % of your current “AI” effort is still reactive? Shift even 10–20% of that energy to proactive and measure avoided incidents, not just closed ones.

86

2 Comments

Today marks Six months in as SOCPAC CTO, and it somehow feels like both yesterday and six years ago. When I stepped into this role, we were behind the curve. Legacy systems, siloed data, and slow processes held us back. In just half a year, we’ve laid a Foundational Technical Architecture that gave us a modern backbone for data, AI, and autonomy. Since then, we’ve: 1. Unified our digital foundation for secure, scalable operations 2. Embedded AI and LLMs into daily workflows to speed decisions 3. Deployed containerized AI agents across all networks 4. Integrated autonomy at the edge, moving toward true man–machine teaming 5. Shifted our culture from cautious adoption to championing innovation The result? We’re no longer catching up, we’re setting the pace. What once felt like a distant goal is now part of our daily reality. All of this happened in six months, which is why it feels so surreal. It took countless late nights, cross-functional sprints, and the relentless dedication of an amazing team. I'm humbled by the trust the organization placed in me to drive this transformation, and I'm grateful to every person who rolled up their sleeves to make it real. This was truly a team effort, from the junior analysts using our new AI tools to the senior leaders who empowered us to take bold risks. And we’re just getting started. The foundation is in place, but now we get to build on it. In the next chapter, we'll be refining and scaling these capabilities, drilling deeper into AI-enhanced decision-making, and pushing the envelope on autonomous operations. The technology will keep evolving, and so will we. Our goal remains the same: give our operators a decisive edge, making it an extension of the warfighter and keep innovating responsibly. I’m humbled by what the team accomplished in such a short time, and excited for what’s next. The foundation is in place. Now we build on it. Six months down, and the journey is just beginning.

170

19 Comments

𝟭𝟬 𝗥𝘂𝗹𝗲𝘀 𝗳𝗼𝗿 𝗮 𝗖𝘂𝘀𝘁𝗼𝗺 𝗖𝗥𝗠 💸 I lost $86K not following these. Don’t do the same. 𝟭/ 𝗕𝘂𝗶𝗹𝗱 𝗳𝗼𝗿 𝗮𝗰𝘁𝗶𝗼𝗻, 𝗻𝗼𝘁 𝗮𝗿𝗰𝗵𝗶𝘃𝗲𝘀 Most CRMs become cluttered databases. Design it to move deals forward. → What do you need to do, not just track? → How fast can someone complete their next task? ✅ Example: → Don’t say: “Status: warm lead” → Say: “Follow up in 2 days — send proposal” 𝟮/ 𝗞𝗲𝗲𝗽 𝗶𝘁 𝘀𝘁𝘂𝗽𝗶𝗱 𝘀𝗶𝗺𝗽𝗹𝗲 You’ll always overbuild. Fight that urge. → 3 core tables: Contacts, Deals, Activity → Automate later, not first ✅ Example: → Don’t add 18 fields to log a call → Just track: Who, What, When, Outcome 𝟯/ 𝗟𝗶𝗻𝗸 𝗿𝗲𝗹𝗮𝘁𝗲𝗱 𝗱𝗮𝘁𝗮 Context is everything. → Every deal links to a contact → Every contact links to a company → Every activity links to both ✅ Example: → Click a deal → see emails, calls, invoices → No tab-hopping to figure out what’s going on 𝟰/ 𝗦𝗲𝘁 𝗿𝗲𝗺𝗶𝗻𝗱𝗲𝗿𝘀, 𝗼𝗿 𝗹𝗼𝘀𝗲 𝗱𝗲𝗮𝗹𝘀 No follow-up = no money. → Add "Next Step" to every deal → Auto-remind reps when nothing’s scheduled ✅ Example: → Don’t say: “In proposal stage” → Say: “Call on Thursday to confirm pricing” 𝟱/ 𝗨𝘀𝗲 𝘃𝗶𝗲𝘄𝘀, 𝗻𝗼𝘁 𝘀𝗲𝗮𝗿𝗰𝗵 Speed = trust in the system. → Create saved views: → “Deals Closing This Week” → “Leads With No Activity” → “Clients Past Due” ✅ Example: → Don’t make reps filter manually → Feed them exactly what they need each morning 𝟲/ 𝗔𝗹𝗶𝗴𝗻 𝗶𝘁 𝘄𝗶𝘁𝗵 𝘆𝗼𝘂𝗿 𝗲𝘅𝗮𝗰𝘁 𝘄𝗼𝗿𝗸𝗳𝗹𝗼𝘄 Templates are generic. Your ops aren’t. → Map: Lead → Deal → Quote → Close → Build just for that flow — nothing else ✅ Example: → Don't use default HubSpot junk → Build Airtable views that match your pipeline 1:1 𝟳/ 𝗔𝘂𝘁𝗼𝗺𝗮𝘁𝗲 𝘁𝗵𝗲 𝗯𝗼𝗿𝗶𝗻𝗴 𝘀𝘁𝘂𝗳𝗳 Your reps shouldn’t be typing “Sent proposal” in 2025. → Automate task creation → Log emails + calls with Make.com → Trigger follow-ups from checkboxes ✅ Example: → Mark deal as “Won” → send welcome email + invoice instantly 𝟴/ 𝗧𝗿𝗮𝗰𝗸 𝗽𝗮𝘆𝗺𝗲𝗻𝘁𝘀 𝗶𝗻𝘀𝗶𝗱𝗲 𝘁𝗵𝗲 𝗖𝗥𝗠 Sales isn’t closed until cash is in. → Add fields for: Deposit, Balance Due, Payment Method → Track outstanding amounts per deal ✅ Example: → Deal says “Closed” → But invoice shows: $500 still unpaid. That’s not closed. 𝟵/ 𝗕𝘂𝗶𝗹𝗱 𝗮 𝗱𝗮𝘀𝗵𝗯𝗼𝗮𝗿𝗱 𝗳𝗼𝗿 𝗮𝗻𝘀𝘄𝗲𝗿𝘀 You don’t want “data.” You want decisions. → Total active deals → Total closed won this month → Sales rep leaderboard ✅ Example: → Use Airtable Interfaces → CEO sees pipeline at a glance — no calls, no spreadsheets 𝟭𝟬/ 𝗧𝗿𝗮𝗶𝗻 𝗹𝗶𝗸𝗲 𝗶𝘁’𝘀 𝗮 𝗿𝗲𝗮𝗹 𝗼𝗻𝗯𝗼𝗮𝗿𝗱𝗶𝗻𝗴 The CRM doesn’t fail. People just never use it. → Loom videos → SOPs in Notion → Weekly cleanup sessions ✅ Example: → Monday morning = CRM hygiene check → “If it’s not in the system, it doesn’t exist Let's Go

35

10 Comments

So which models produce HTML that has the least accessibility errors, and at what cost? Mapped it out here (bottom left is the best). This is helpful for understanding which models are most compliant for those with disabilities and do it efficiently. Qwen and Gemini 2.5 Flash lead the pack. Hat tip to Ben Ogilvie for pointing me to this: aimac.ai

50

2 Comments

What are MCPs and why should AI Operators pay attention? APIs were how we connected systems. MCPs are how we’ll connect agents. If you’re new to MCPs - think of them like a smarter API. Instead of needing exact instructions, you can tell it what you want (“a 4-course pasta dinner”) … and it figures out the best way to make it happen. That’s what makes them powerful for AI Operators. If you already run your work through a clear playbook, MCPs just give you a smarter way to run a step. It's not a fancy new game-changing you-must-drop-everything-and-switch-to-this tool like social media may make you think Just a faster way for your agents to do the work you define. What's nice is that they are what we call "modular" - meaning once the MCP is built, any other agent can plug into it and run instantly. Think of it like giving your agents a shared “language” for executing your playbooks. Have you dabbled in MCPs yet? What questions do you have about them?

45

12 Comments

How to give engineers requirements in 2025: (Hint: it’s not a ticket that says “Add a search bar”) Start with the why: “Our users are frustrated because they can’t find anything once their data grows past a few dozen items. Support tickets for this jumped 40% last month.” Define the scope: “This needs to work for both mobile and desktop. We’re not optimizing for advanced filters yet, just basic keyword lookup across titles and tags.” Give clarity: “The goal is to reduce time-to-discovery. Success means users can find what they need in under 5 seconds. We’ll measure success through session recordings and search success rate.” Reinforce focus: "Speed and simplicity matter most. We’d rather ship a fast, usable v1 than a perfect UX.” Then step back. Let your engineers own the how. Not just build what was asked. Because real requirements in 2025 aren’t just specs. They’re context. They’re constraints. They’re customer outcomes. That’s what it means to be Product Driven and that's exactly what we do at Algorithm

51

11 Comments

Jina finding a home in Elastic marks the third acquisition/pivot in this space just this year. Many of these «embedding model» companies have presented themselves as that, but in practice have been selling «enterprise search» and consulting for years. Offering managed endpoints for embedding inference wasn’t a good business model after the initial post-chat moment (See link for a great read on that) Jina acquired by Elastic, Voyage acquired by Mongo, and Mixedbread pivoted to managed no-code search endpoints. Everything changes https://lnkd.in/dv7wx9-b

111

10 Comments

What's the MACH Alliance for? It's a question I've been asked a great deal and to take it back to first principals it was to promote architectures (unashamedly technical) to support customer choice, excellent user experience and interoperability between best of breed vendors. Has that worked? That's a debate that can last hours - the good MACH projects, the ones on stage at the various MACH events show that it absolutely can work. Have there been horror stories - of course, plus ça change in IT. However, one thing that has been announced and now released is the MACH Common Data Model. Search my LinkedIn and you'll find a link to a (now unavailable) blog I wrote in 2019 (I think; Stacey Trimnell-Ritchard remembers it I hope as she said she used it in a marketing conversation back then so it did happen 🤣) calling for this and so it's brilliant to see it now exists: https://lnkd.in/eTFVgYKX What is it? In their own words: "The MACH Alliance Open Data Model (ODM) initiative is the natural next step in enabling interoperability across composable ecosystems. It serves as a semantic interoperability layer: a "Rosetta Stone" for understanding how to translate data between systems in hybrid commerce architectures. It provides canonical reference models and implementation guidance that enable architects to design effective data integration across commerce platforms while preparing organizations for AI-powered ecosystems." I'm interested and signed up and I know we're looking at how we can incorporate the ideas into Patchworks (the platform already supports this of course and it will help our AI mapping tools no end). Find out more by attending MACH X in October: https://lnkd.in/epHx2JVp

42

3 Comments

SalesLoft just announced they got breached through OAuth token theft. (https://lnkd.in/gdr4v3Rt) I think this is the canary in the coal mine for what's likely to become a massive problem. OAuth tokens are becoming the path of least resistance for attackers. Why spend months trying to breach Google, Salesforce, Shopify, Atlassian... (enter any SaaS app here) directly when you can compromise one small vendor that has OAuth access to thousands of accounts? The math is pretty straightforward - hack one app, get access to thousands of customer instances. That's an ROI any attacker would take. Pat Opet, CISO at JPMorgan, recently published an open letter that really resonated with me. He essentially said that third-party SaaS risk is one of their biggest concerns. When JPMorgan's CISO is publicly worried about this, I think we all need to pay attention. (Here's his letter: https://lnkd.in/g3vEAyZG) Here's something interesting - go to https://lnkd.in/g59W2D6N and look at all the apps that have access to your Google data. I did this and found apps I hadn't used in years. Some I didn't even remember installing. Many still had active access to my email, calendar, and documents. It's eye-opening, and honestly, a bit concerning. What Companies Should Be Doing? First, audit your OAuth grants regularly. Check every SaaS app that connects third-party apps. I guarantee you'll find surprises. Apps from former employees, tools nobody remembers approving, permissions that don't make sense. Second, actually vet vendors before connecting them. I know we all ask for SOC 2 reports, but that's not enough anymore. We need to ask harder questions like: What happens when you get breached? How do you store our OAuth tokens? Can you detect if they're being abused? Third, disconnect what you're not using. Those old OAuth connections are like leaving doors unlocked. There are actually some good tools now - companies like Nudge Security that can automatically monitor and manage OAuth sprawl. If you're not using something like this, you're basically flying blind. Fourth, and this is where my bias shows - backup your data! When someone uses legitimate OAuth tokens, your security tools just see normal API traffic. You won't know something's wrong until your data is gone. The ability to restore quickly might be your only real defense. The Bigger Picture I think we've all been so focused on the productivity gains from SaaS (which are real and substantial) that we haven't properly considered the security implications. Every OAuth token is essentially a permanent password to your data that you've handed to a third party. The companies affected by the SalesLoft breach are learning this the hard way. Their data was compromised not because their own security failed, but because they trusted someone else's.

34

4 Comments