Joe LeBlanc

I've been spending a lot of time experimenting with agent frameworks lately and testing them in real workflows, and I keep seeing the same issues show up repeatedly. Curious if others are running into the same things. Some of the most common failure patterns I'm seeing: 1. Agents making random tool decisions Even with structured prompts, the model sometimes selects the wrong tool or workflow step. It works most of the time, but every once in a while it makes a strange decision that sends the workflow down the wrong path. 2. Workflow loops Agents can get stuck repeating steps or retrying actions that already failed. In long chains this can escalate quickly. 3. Exploding API costs A lot of orchestration logic ends up being handled by the LLM itself. When the model is used for routing decisions, tool selection, and reasoning, the number of calls multiplies very fast. 4. Non-deterministic behavior The same task can produce slightly different execution paths depending on wording or minor context differences. That makes debugging difficult. 5. Safety boundaries Once agents start interacting with tools or data systems, controlling what actions are allowed becomes tricky. What I'm starting to suspect is that many agent architectures rely on probabilistic models for decisions that might be better handled deterministically. LLMs are incredible for reasoning and generation, but I'm not convinced they should always be responsible for task arbitration and workflow routing. Curious how others here are approaching this. Are people solving these problems with: • better prompting • guardrails • tool restrictions • workflow engines • something else entirely? Would love to hear what solutions people have found that actually work in production.

When we hire agencies for work we do not know how to do, I often feel the absence of a collaborative stance. I am learning that many agencies guard separation because their value story depends on being the experts who deliver. They hold the boundary because permeability can blur that identity. I see now that the deeper issue is not the tool or the workflow. It is the organization’s identity and how it relates to co-creation. If the identity is built on delivery, then collaboration feels like dilution. My responsibility is to state the requirement for collaboration early. It must be part of both design and delivery. When this is clear at the start, we avoid the confusion that comes when the work asks for openness that the relationship was not built to support. [December 16 Update] This is applicable both externally and internally. Many S&F client relationships still haven't evolved beyond the expert delivery paradigm.

27

5 Comments

I use Claude for client code reviews now instead of hiring a senior dev. Here's the process: 1. Paste their code 2. Give context: "We're using Rails 8 with Hotwire. This is a new feature for subscription management." 3. Ask: "What's inefficient, what's dangerous, what could break?" It catches things my tired brain misses at 5pm. Security issues. N+1 queries. Over-engineered solutions. Then my dev and I discuss the feedback together. Cost: $0.15 per review Time saved: 2 hours of my review time Quality improvement: Actually better feedback than some senior devs give Not replacing people. Augmenting them. This is what 2026 AI looks like for small teams: strategic leverage, not replacement. What's one expensive process in your business you could audit with AI first? #AI #Claude #SoftwareDevelopment #CodeReview #TeamEfficiency #Rails #Productivity #Consulting

3

7 Comments

Do you buying technical debt? If your agency just takes your project and passes it to someone else, you’re buying tech debt you’ll pay for later. Real developers don’t just code, they solve problems, debug, and care about your business. Middlemen? They sell promises, not quality. Your app might work for 10 users… but crash when you hit 1,000 ( like as shows in this image ) Want a product that scales, works under pressure, and actually solves problems? Go straight to developers who care. Stop paying for people who sell projects. Start paying for people who deliver solutions 🌟 . #SaaS #MVP #Startup #Founders #ProductDevelopment

1

One attendee told me this topic should be at every tech conference. Was I proud? Yes. But also a bit concerned… A couple weeks ago I spoke at an amazing LeadDev LDX3 conference about something that’s been on my mind for 5 years: why so many engineering managers are burning out. 🔥🔥🔥 That's why I gave a "Don't Get Toasted: A Leader's Guide to Mental Well-Being." talk. I'll be honest - I was more nervous about this presentation than any technical talk I've ever given. Would people think discussing mental health at a tech conference is too soft? Would they even care? Turns out, they really cared. After the talk, multiple people told me "This resonates so much," "We need this conversation," and "This should be at every tech conference." Here's what struck me: we've built an industry where we obsess over system architecture, code reviews, and deployment pipelines (as we should!). But we rarely talk about the mental state of the humans running these systems. For last couple of years I've noticed that management work is becoming increasingly psychological. We handle difficult conversations, AI hype pressure, layoffs, personal issues... yet most of us have zero training for the emotional aspects of supporting others. Maybe it's time we treat leaders' mental well-being as seriously as we treat system reliability? When an engineering manager is burned out, the whole team feels it. What's your experience? Have you seen mental health topics at tech conferences you've attended? What's missing from leadership conversations in our industry? Link to the full talk recording in the comments below 👇 #TechLeadership #EngineeringManagement #MentalHealthAtWork #LeadershipDevelopment

48

13 Comments

Now that NPS allows up to 100% equity allocation, let’s go all-in on equity. A 40-yo software developer texted me this at midnight, buzzing with excitement. When I called him the next day, his enthusiasm was overflowing. “Sameer, do you even know your current equity allocation?” I asked. “Does it matter?” he shot back. “Yes, it does. Equity is equity… whether you invest via stocks, mutual funds, or NPS. Your equity allocation is already 65%, and pushing NPS to 100% equity will expose you to unnecessary long-term risk. Let’s stick with our current allocation strategy.” It was tough to convince him, but he trusted me and dropped the idea. But this may not happen with you, if you do DIY investing, or your advisor doesn’t look at two key parameters: 1) Risk profile 100% equity suits only those with an aggressive risk appetite. If you’re undergoing a major life event… marriage, kids, loans… you may not have the stomach to face excess volatility. 2) Asset allocation strategy Equity is equity. Irrespective of the asset class. If you’re already heavy on equity, any further NPS allocation can put you in deep risk. And don’t forget valuations. Going all-in at market highs is a big risk in itself. In my early career, a mentor told me: “Finance will always throw shiny new options at you. But a successful investor is like Arjuna… eyes fixed on the target, not the distractions.” I’ve lived by that advice. You should, too.

48

13 Comments

Reporting on hackathon outcomes is hard. Usually people use participation numbers, but it's business outcomes that matter as much or more. Devpost for Teams has five primary features to help report on ROI: 1️⃣ Custom registration questions to gather important info on each participant. -- "Have you used these tools before?" "What do you hope to gain out of this hackathon?" and more. 2️⃣ Project descriptions with built-in prompts -- "What problem are you solving and why?" These get to the value of the projects themselves. 3️⃣ Custom submission questions -- These further detail the outcomes, and can be made 'hidden' from the viewable project if desired. "How much time will this save you per week?" "How extensively does this address a key business goal?" "What more needs to be done to get it to production?" 4️⃣ Post-hackathon survey -- To circle back from the project to the participant and compare this hackathon to others. "Did this hackathon strengthen collaboration and teamwork for you at the company?" "How likely are you to apply the skills or knowledge gained from this hackathon to your future work?" 5️⃣ Judging, voting, and feedback -- What were the responses from the judges and colleagues, to help gauge and crowdsource quality? All of the above, plus basic metrics (registrations, submissions, team make-up, Geo) and our services team, help report on hackathon ROI to stakeholders. Without it, you're just running an event. Here are two blog posts that dive deeper: → "Streamline feedback with Devpost for Teams’ Post-Hackathon Surveys" https://bit.ly/45wsM3J → "Measuring the success of your internal innovation hackathon" https://bit.ly/4n0qa4m #hackathons #ROI #AI

18

The longer your hiring process, the worse the hires. We see it constantly at Remotely.works, and the rigor is almost never what explains the difference between the ones getting it right and the teams getting it wrong. So I mapped what's actually working in developer hiring against what just looks like it is. Especially now that teams are hiring fewer, more senior engineers and the old playbook wasn't built for it. Most hiring budgets are going into the top-left of this chart. Process-heavy screening, panel interviews, AI sourcing with nobody actually reading the output. It looks rigorous and I think that's mostly the point. It doesn't produce better hires. The bottom-right is where I'd start if I were hiring engineers tomorrow. Transparent pricing, faster feedback, shorter loops. None of it is hard. Most teams just haven't gotten around to it because the old way still works well enough. The top-right is harder and it's where we built Remotely. Engineering-led hiring, AI-assisted technical assessments, screening for AI judgment, vetting for ownership, and blended teams. It takes longer to set up and most companies won't put in the work, which is honestly fine with us. If you're still running the left side of this in 2026, the problem is that the industry moved and you didn't.

12

Production AI architecture is starting to feel familiar. Love watching architecture patterns crystallize in real time. Latest example: Anthropic will bring an advisor/executor pattern to the Claude Platform. Opus as the strategic brain, Sonnet or Haiku doing the heavy lifting. Smart abstraction. Martin Fowler wrote about layering principles decades ago. Turns out that principle doesn't stop at software layers. Now one of the layers is a frontier model and the other is a slightly-less-frontier model 😉. Some of these patterns borrowed from distributed systems, some from classic OOP, some brand new. Exciting time to be paying attention 👀.

6

A team member messaged me after a 4-hour call fixing dev issues. "I can't believe you were on that call." We have 100+ people at Genflow. I run a $100M agency. And I was right there debugging code with the team. It hit me how wrong most people get leadership. Everyone thinks being CEO means strategy sessions and vision boards. Wrong. Being CEO means being the captain who's still on the field. You show people how it's done by doing it alongside them. Not above them. The idea that you'll build a business and just work on "high-level stuff" while everyone else executes? That's not how you scale. You scale by staying close to the work. Understanding the actual problems. Building alongside your team. After 9 years of doing this, I've learned: The best CEOs aren't in corner offices. They're in the trenches when things break. They know the product inside out. They can still do the work if needed. Your team doesn't follow titles. They follow leaders who roll up their sleeves. What kind of leader are you trying to be?

44

16 Comments

I've talked to hundreds of founders spending 40+ hours a month on ops work that shouldn't be theirs. Writing briefs, scoping roles, explaining the same task twice. So I put together 10 Claude prompts to remove that layer completely. → One-line task into a contractor brief → Slack message into a clean handoff doc → Vet a candidate portfolio in under 10 minutes → Scope a role without a job description → Audit your week and cut what shouldn't be yours Copy, paste, and move on with no setup needed. Completely free. Comment "PROMPTS" and connect with me. I'll DM it over.

8

6 Comments

Asking 5-Whys is a lie. Every "Why?" spawns several more questions. It's not a list. It's a tree. As an engineering leader, your job is to make sure your team prunes the tree. That they don't get lost in endless shallow questions. It's to push them to go deep. Not just anywhere, but where it matters. This is just one of the ways that you can bring value to your team after an incident. I cover this more in depth in the latest δeltas newsletter: https://lnkd.in/eqpE_YeV

11

2 Comments

The most frustrating part about helping experienced developers find jobs? Watching them apologize for wanting work-life balance. "I know this might sound unreasonable, but..." "I hope it's okay that I'm not in that stage of life anymore...." "I'm sorry, but I need to pick up my kids at 5..." Stop. Just stop. You don't need to apologize for having a life. You don't need to apologize for setting boundaries. You don't need to apologize for wanting to see your family. The companies worth working for understand this. They know that: - Rested developers write better code - Happy developers stay longer - Balanced developers are more creative - Experienced developers deliver quality over quantity - Paying 30% more results in 100% better output I've seen too many talented developers burn out because they thought grinding was the only way to prove their worth. Here's what I tell every developer I work with: Your value isn't measured by the hours you log. It's measured by the problems you solve. The systems you build. The knowledge you share. The mentorship you provide. A company that can't see that isn't worth your time. Period. Your boundaries aren't negotiable. Your experience is valuable. Your time matters. Don't let anyone convince you otherwise.

20

6 Comments

Stop paying a $150/hour engineer to write documentation. Save this for the next time someone says, “Our devs will just handle the docs.” Your codebase is a Ferrari, but the user guide you asked the backend team to slap together is a tricycle. Here’s what actually happens when devs moonlight as tech writers: 1️⃣ 10–12 hours a week vanish: Devs already lose up to 10+ hours hunting for information. Pile on doc duty and you’re compounding waste. 2️⃣ Support tickets spike: Rush jobs miss edge cases, screenshots go stale, onboarding questions flood Slack. Your “savings” turn into Level-1 support payroll. 3️⃣ Feature velocity stalls: Every minute in Google Docs is a minute not shipping the next release. That delay hits revenue, not just morale. What you really need is an engineer-led documentation agency. Here’s the baseline when we take doc duty off the dev team’s plate: • Onboarding time drops 80%. • Support tickets drop 90%. • Developers get a full day a week back to build. Documentation is a specialist craft. Treat it like design, product, or QA: dedicated owners, repeatable SOPs, measurable ROI. Let developers develop. Let experts document. If you’re done burning senior-engineer hours on work that isn’t moving your product forward, comment ‘Doc Duty’ below for a free strategy call and my documentation start-up playbook. PS: Make sure we’re connected first so I can send it over.

16

I know I’m going to get heat for this from candidates, BUT I genuinely believe every interview process—technical or not—should include some sort of assessment. Live coding for technical roles, take-home presentations for design/product, a pitch or case study for GTM, People, and Finance roles. Why? Because it’s not just about proving your skills—it’s also a great way for you to understand what the actual work will be like. (Assuming the company puts together a reasonable case study... and doesn’t make you pitch their product for free 🤣 ). If done right, it’s a win-win.

27

30 Comments

Most non-technical software founders don’t fail because they picked the wrong idea. They fail because they stayed with the wrong scrappy development setup for too long. I’m seeing more non-technical SaaS founders ship something quickly than ever before. A few AI tools, a scrappy contractor, maybe an offshore team, and you’ve got a working product in weeks. That early speed feels like momentum. But working software is not the same thing as scalable software that can be the core of your business--and core to your customers' workflows. And most founders don’t know the difference until growth exposes it. Slowly, over months, the product performance lags, features take twice as long to add, security questions get uncomfortable, and no one can clearly explain how the system actually works. 🔧 Keith Shields 🖌️ , who leads Designli, has spent over a decade helping industry experts turn their domain knowledge into real SaaS products. He built his firm after dealing with unreliable development partners himself, and now runs a team that effectively serves as a full-stack outsourced engineering department for vertical SaaS founders. What stands out in our conversation isn’t just a better process or AI tools for making software. It’s this: new founders wait too long to get an outside technical perspective. They feel stuck. Or hopeful. Or too invested to question the current path. But a simple third-party review of your codebase and development process can surface risks early—before you commit another year and another million dollars in the wrong direction. Serious founders don’t just validate product-market fit. They validate technical foundations. Keith tells this better than I just did. Have you ever realized too late that your first development setup wasn’t built for the company you wanted to become?

33

14 Comments

Newly-hired CROs: Don't just believe them when they tell you what your job is. Yes, the last CRO was fired because things weren't going well. But that doesn't mean the team knows what the real problem was. Here I talk with Kyle from Pavilion about a time I learned that lesson firsthand. 📺 https://lnkd.in/gVS9MeJf 📖 mybook.to/CROsGuide

4

But yes you should fire all your devs and become reliant on some 3rd party company to spew out code you don’t understand. What could possibly go wrong? Aside from this? Maybe they decide one day they actually have to make money. When your subscription suddenly goes up several orders of magnitude? Hope you have a business plan for that.

2

I stopped building "lucky" automations 6 months ago. I used to build workflows that only worked when everything was perfect. One website update. One API change. One UI shift. Everything would crash. I learned this the hard way during a live demo for a $3,000 client. My n8n scraper hit a changed CSS selector and just stopped. I was embarrassed. I spent 4 hours that night manually fixing code I thought was "automated." That is when I discovered the Principle of Anti-Fragility. In my Computer Engineering days, we learned that systems should get stronger under stress. Now, I don't just use n8n. I wrap my workflows in an agentic layer using Claude Code and OpenClaw. Here is the exact growth hack for a self-healing system: 1. Error Handling: Add an 'Error Trigger' node in n8n. 2. Vision: When a scraper fails, it calls OpenClaw to take a screenshot of the page. 3. Repair: The screenshot and the broken code are sent to Claude Code via a CLI command. 4. Deployment: Claude Code identifies the new UI element, rewrites the selector, and pushes the fix back to the n8n function node. Last week, a client's lead source changed their entire dashboard. The workflow broke at 3 AM. The agentic layer saw the error, used OpenClaw to find the new "Submit" button, and Claude Code updated the script. By 3:01 AM, the system was running again. I didn't even wake up. Stop building fragile workflows. Start building agents that fix themselves. Is your automation "lucky" or is it resilient? Drop a comment below. Enjoy this? ♻️ Repost it to your network and follow Jadai Kongolo for more.

2

1 Comment