About
Outcome driven software engineer with expertise in building AI agents and Machine…
Activity
2K followers
-
Matthew Feroz shared thisI filmed an episode of AI Agent Clinic with Google Cloud! We built an automated eval suite for a docs agent I built in LangChain, using Merge Gateway as the intelligence provider. Shoutout to the three OSS projects I ran my agent on: • T3 Code by Theo Browne + Julius Marminge • OpenCode • Pi by Mario Zechner Huge thanks to Dani Zamora for having me on!Matthew Feroz shared thisIf your testing strategy for AI agents consists of running 3 manual prompts in your terminal and saying “looks good to me,” you don’t have an agent ready for production—you have a prototype. The hardest part about autonomous loops (like LangGraph or CrewAI) isn’t hard crashes—it’s silent failure. Agents will execute without throwing a single 500 error while quietly hallucinating or generating subpar outputs. In episode 3 of the AI Agent Clinic, Dani Zamora sits down with Matthew Feroz, Developer Advocate at Merge, to upgrade his DocsHound agent in 60 minutes. At first, you'll see that Matt was confident in his agent’s output. But once we hooked up an automated evaluation pipeline, the data told a different story: a 33% quality score on documentation accuracy—a complete blind spot that manual testing never caught. In this episode, we break down the 4-step framework to evaluate any AI agent: 1️⃣ Map Execution Flow: Pointing coding agents to source code to inspect inner workings. 2️⃣ Standardize Telemetry: Using OpenTelemetry and OpenInference so your eval toolset works across any framework. 3️⃣ Define Quality Rubrics: Turning subjective developer expectations into structured LLM-as-a-judge metrics. 4️⃣ Visualize Direction: Running scorecards to spot exact regressions in latency, token cost, and accuracy. Check out the full 60-minute build → https://goo.gle/4hHEXiO
-
Matthew Feroz shared thisLast night, I spoke at the AI Nerd Meetup in NYC alongside Fireworks AI, Harmonic, and Nori Agentic. I shared why owning your intelligence matters in the AI era, and how we’re approaching that at Merge. We handle the connectivity layer so teams can focus on building their product instead of rebuilding integrations. Thanks to everyone who came out!
-
Matthew Feroz reposted thisMatthew Feroz reposted thisIntroducing Merge's Universal Context Layer. It substantially decreases token cost while increasing accuracy. Every company uses AI wrong: Say your designer makes a deck using Claude. They teach Claude your design systems and typography. Now a Marketing hire makes a deck. None of the information your designer taught Claude gets passed over… why? Merge fixes this: All knowledge and memory across your entire company gets stored inside our context layer. When someone on the org teaches Al, that skill lives across the entire organization instead of staying locked in their session. Your AI runs faster, cheaper, and with more accuracy. You don’t have to switch systems, incur more costs, or retrain anyone on your team, Merge runs invisibly in the background. Book a demo here: https://lnkd.in/gH-8wWGp
-
Matthew Feroz shared thisThis is such an insane deal. GLM 5.3 Flash is one of the smartest, fastest, and cheapest models right now, try it today on Gateway!Matthew Feroz shared thisUntil the end of September, GLM 5.3 Flash costs less than a cent to run a task. That price tier used to top out at 39 out of 100 on the Artificial Analysis Intelligence Index. GLM 5.3 Flash scores 57. It's 90% off in Merge Gateway: $0.012 per million input tokens, $0.04 output, $0.003 cached.
-
Matthew Feroz shared thisHappy to announce that last night I hosted my first ever Merge tech event in NYC with PostHog and Redis! We packed a room to hear how teams are building autonomous AI systems that can use context, take action, and improve over time. Thanks to our speakers Tim Bourcier, Marco Gancitano, Nitin Kanukolanu, Shefali Parmar, Ben Danzig, everyone who joined us, and Kyle Tsang for this sick edit!
-
Matthew Feroz shared thisLast week I hosted a 200 person company wide hackathon for Merge! The theme was Pirate Party if you couldn't tell 🦜 I feel incredibly lucky to bring what I’ve learned from hosting events across NYC into my work at Merge. Huge thanks Katie Flattum Magnuson for co-planning alongside me! Couldn't have done it without you.
-
Matthew Feroz shared thisevents events eventsMatthew Feroz shared this
-
Matthew Feroz shared thisSo hyped to be hosting this event alongside PostHog 🦔 !!!Matthew Feroz shared thisOn August 27, Merge and PostHog are hosting The Self-Driving Product, an evening of technical talks and live demos for NYC builders. We’ll get practical about observability, evaluations, and the infrastructure required to have your product drive itself. Building something that fits the theme? Apply to demo it live during our community showcase! RSVP and apply at https://luma.com/gfn0jy68
-
Matthew Feroz shared thisYesterday we launched the Merge Startup Program! We're providing thousands of FREE credits for early-stage teams to start building. Try it out: https://lnkd.in/gWpk4b6K
Matthew Feroz commented on a post
2w
yooo huge dub
-
Matthew Feroz liked thisAfter a layoff from Redis in July, I took my first real break in almost a decade. It gave me time to reflect, learn, and be intentional about what comes next. I’m excited to share that I’m in my third week at OpenAI where I’ve joined the GTM Enablement team. It brings together my passion for technology and learning. I’m grateful for everyone who helped me get here, and to Mimi Cai, Leah Conner, Brooke Bachesta, Ryan A., and so many new teammates who have already made me feel welcome. The timing made yesterday’s DevDay especially thrilling! I’m already experimenting with Dots to help me learn and work differently: connecting information, finding resources, brainstorming, and building. There’s a lot to explore! I’m also incredibly thankful to Amanda Gaube, Megan Viazanko, Kristine P., Joe C., Steve Jenner, and so many others at Redis for a great year working on Solutions Architecture, business strategy and enablement. So much is possible right now. I’m looking forward to contributing and helping more people put AI to work in meaningful ways. Check out what OpenAI launched at DevDay 👇 #OpenAI #DevDay #AI #EnablementMatthew Feroz liked thisFrom personal agents and collaborative work spaces to a new affordable frontier model and an ultrafast mode—here’s a peek at what just dropped at DevDay 2026 ⬇️ https://lnkd.in/e4jfz3dm
-
Matthew Feroz liked thisMatthew Feroz liked thisIf your testing strategy for AI agents consists of running 3 manual prompts in your terminal and saying “looks good to me,” you don’t have an agent ready for production—you have a prototype. The hardest part about autonomous loops (like LangGraph or CrewAI) isn’t hard crashes—it’s silent failure. Agents will execute without throwing a single 500 error while quietly hallucinating or generating subpar outputs. In episode 3 of the AI Agent Clinic, Dani Zamora sits down with Matthew Feroz, Developer Advocate at Merge, to upgrade his DocsHound agent in 60 minutes. At first, you'll see that Matt was confident in his agent’s output. But once we hooked up an automated evaluation pipeline, the data told a different story: a 33% quality score on documentation accuracy—a complete blind spot that manual testing never caught. In this episode, we break down the 4-step framework to evaluate any AI agent: 1️⃣ Map Execution Flow: Pointing coding agents to source code to inspect inner workings. 2️⃣ Standardize Telemetry: Using OpenTelemetry and OpenInference so your eval toolset works across any framework. 3️⃣ Define Quality Rubrics: Turning subjective developer expectations into structured LLM-as-a-judge metrics. 4️⃣ Visualize Direction: Running scorecards to spot exact regressions in latency, token cost, and accuracy. Check out the full 60-minute build → https://goo.gle/4hHEXiO
-
Matthew Feroz liked thisMatthew Feroz liked thisSending every agent task to the most capable model gets expensive fast. Eragon now routes each task to the right model through one Merge Gateway endpoint. Customers get the same quality, while Eragon saves hundreds of thousands of dollars in inference. Full case study in comments.
-
Matthew Feroz liked thisMatthew Feroz liked this
-
Matthew Feroz liked thisMatthew Feroz liked thisSeptember flew by at Merge. Between the billboards, Lenny & Friends Summit, and 8+ dinners, we still found time to ship: Universal Context Layer, new models on Gateway, Batch Requests, and Self-Hosted Models. Stay tuned for what we've got cooking in October.
-
Matthew Feroz liked thisMatthew Feroz liked thisThe best open-weight model for code is third overall, at a 23rd of the price of the model above it. MiMo-V2.6-Pro scores 60.9 on SciCode against Claude Opus 5.5's 66.9, and runs $0.87 per million output tokens against $20. Kimi K3 and GLM-5.3 follow within two points of it. Six points for $19 per million tokens is a real tradeoff, and it's the kind a Build-Your-Own-Router policy on Merge Gateway lets you set per request rather than once. Scores via artificialanalysis.ai, week of Sep 28.
-
Matthew Feroz liked thisMatthew Feroz liked thisThe $200 AI buffet is closing. Agents broke the subscription model. When one power user can burn thousands of dollars of API-equivalent compute on a $200 plan, “unlimited” stops working. What comes next is token surge pricing. Anthropic’s path to the public markets helps explain why... Private markets can subsidize compute economics. Public markets demand better margins. And the shift is already starting: --> Peak-hour limits. --> Session + weekly caps. --> Overflow at API rates. --> Premium pricing for speed, priority, and long context. --> Models charging tens of dollars per million output tokens. That’s Uber economics. You get the included ride. Then the meter starts. OpenAI is moving the same way: the sticker price can stay at $200 while the amount of compute included underneath it gets tighter because agents changed the math. A normal chat is cheap. A coding or research loop can consume 100–1,000× more tokens. That works when most subscribers barely use their allowance. It gets much harder to defend when investors start obsessing over gross margin. So AI pricing likely becomes three layers: 1. Smaller included quota 2. Higher pricing when compute is scarce 3. API-style billing once you blow through the bundle The heaviest users stop being the loss leader. They become the business. Tokens were priced like electricity in a village. Surge pricing is what happens when the village IPOs.
-
Matthew Feroz liked thisMatthew Feroz liked thiswhat's happening on the ground: partnerships edition 🌻🫶 first few weeks at Hera and it's been a whirlwind. the highlights: ACMA NY Chapter! my first conference with this team + my first ACMA (many more to come). we had a small activation and asked everyone to share something they'll always remember about an adult who shaped their life. mine is my grandma: unbelievably selfless, shared everything, even her favorites. also hated getting her blood sugar checked lol. loved seeing familiar faces from our hospital partners and meeting the people on the floor a group salsa class! the team so graciously joined me for a group salsa class led by my favorite instructor. it is very clear Jenny was a professional dancer in her past life our office cat, Toro! he stops by every now and then and it's always a treat when Connie brings him in healthcare has been a learning curve, but I'm learning from incredible people. a team of unbelievably sharp heroes (care managers) who live and breathe this every day. the best part is spending my days with the people in the health systems we partner with and hearing how much we've been helping their teams and patients 📍more conferences and events coming up. who's the adult you'll always remember? drop it below or message me #healthcare #partnerships Isabella Petas, Ellen Dong, Sasi Kota, Kenny Derek, Duncan McManus, Myles Novick, Will Sullivan, Connie Kang, Jenny Lee, Yaakov Friedman, Tatiana Lloyd-Dotta, LMSW, Jacinda Schell, Shaneka Swaby, MSW
Experience & Education
Licenses & Certifications
Courses
-
Agile Methods for Software Dev.
SSW 555
-
Software Architecture & Component Based Design
SSW 565
Other 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