💥 Issue 94: How to hack the job market, AI won’t fix developer productivity, Message delivery in distributed systems, Clean Architecture

The silent infrastructure crisis, Whats New with Amazon OpenSearch, Clean Architecture with Python

Feb 18, 2026

You forward conference talks to your team. What if you watched them together and talked about it after? (Sponsored)

Your team is navigating the same decisions everyone else is. How do we evolve our architecture without stopping the business? What does production AI actually look like? Where should we invest our time?

At QCon London (March 16-18), senior engineers from Netflix, LinkedIn, and Goldman Sachs share how they worked through it. Your team learns it together, debates it together, and leaves aligned on what to do next.

Save your team’s spot. Group pricing for 3+.

Save your team's spot.

Happy Wednesday 👋 and a warm welcome to Tech Talks Weekly #94!

Last week, I published a compilation of the most-watched Java talks of 2025 which went viral. Make sure to check it out:

🏆 100 Most Watched Java Conference Talks Of 2025

Tech Talks Weekly

Feb 10

Read full story

This issue includes over 65 new conference talks published in the past week together with 85 new podcasts. That sounds like a lot, so make sure to check out both 🏆 Featured this week as well as 📈 Most-watched talks to find the ones you don’t want to miss.

Get ready and let’s jump right in!

🏆 Featured this week

Here are our top recommendations this week. Watch now or bookmark!

“Learning not leaning: How developers can Accelerate their Learning with AI - Emma Burstow” from NDC London 2026
Conference ⸱ +1k views ⸱ Feb 11, 2026 ⸱ 00h 55m 44s
tldw: Coding assistants feel like the next Intellisense, but they change how people learn, not just how they type. The talk argues the goal is not more code output, it is becoming a better engineer by using AI as a learning tool. A team tried structured Copilot sprints where everyone had to use it, but prompts had to focus on why and what if, tradeoffs, and alternative designs instead of code dumps. Treating the assistant like a tutor, asking for explanations, tests, edge cases, and reasoning, helped developers practice and make mistakes safely while still shipping fast.
“Ensuring reliability: message delivery in distributed system • Michał Ostruszka • Devoxx Poland 2024” from Devoxx Poland 2024
Conference ⸱ <100 views ⸱ Feb 18, 2026 ⸱ 00h 50m 14s
tldw: Distributed systems drop, duplicate, and reorder messages all the time, so “message delivered” is rarely a simple yes or no. This talk explains common delivery guarantees like at most once, at least once, and exactly once, what they actually mean in production, and when you should not bother chasing stronger guarantees. It walks through techniques like idempotent handlers, deduplication keys, retries with backoff, acknowledgments, outbox and inbox patterns, and how brokers and databases change the tradeoffs. The main takeaway: pick the weakest guarantee that meets the business need, then design the system so failures turn into safe retries instead of bad state.
“AI won’t fix developer productivity (unless you fix context first) | Peter Werry | StaffPlus NY 2025” from StaffPlus New York 2025
Conference ⸱ +200 views ⸱ Feb 13, 2026 ⸱ 00h 22m 59s
tldw: AI coding tools are not boosting productivity for most teams because the real bottleneck is context, not code writing. Developers spend tons of time hunting through GitHub, Slack, Jira, and Confluence to figure out why something exists and what tradeoffs were made. The talk argues for “context engineering”: building a layer that links decisions, discussions, and system history so both humans and tools like Claude or Cursor can query it directly. Done well, this cuts guessing, speeds up changes, and makes AI outputs more reliable for real production codebases.
“The silent infrastructure crisis | Chris Buckley | StaffPlus New York 2025” from StaffPlus New York 2025
Conference ⸱ <100 views ⸱ Feb 11, 2026 ⸱ 00h 13m 26s
tldw: AI crawlers are turning into a new kind of production traffic. Data from Fastly’s sample of 6.5T monthly requests shows single “fetcher” bots spiking as high as 39,000 requests per minute, creating load patterns that don’t look like humans and don’t behave like classic scrapers. The talk argues this is now an infra and security problem: more cache misses, more origin hits, more expensive egress, noisier telemetry, and tougher rate limiting decisions since some bot traffic is “legit” and some is abuse. This talk includes concrete suggestions for how to detect, classify, throttle, and price this traffic so it doesn’t quietly eat your performance budget.
“Whats New with Amazon OpenSearch Service” from AWS Podcast
Podcast ⸱ Feb 17, 2026 ⸱ 01h 00m 29s
tldl: Amazon OpenSearch Service got a batch of updates aimed at people running search and analytics in production. The talk walks through new observability features, better vector search for embedding based queries, and performance work meant to make indexing and querying faster and cheaper at scale. It presents the changes around day to day ops: seeing what your cluster is doing, improving relevance for semantic search, and getting more throughput from the same hardware.
Do you like the issue? Hit the like or leave a comment. Thank you 🙏
“Clean Architecture with Python • Sam Keen & Max Kirchoff”
Podcast ⸱ Feb 13, 2026 ⸱ 00h 36m 56s
tldl: This interview talks through applying Clean Architecture ideas in Python. It focuses on keeping business logic separate from frameworks and I/O code so you can test core behavior fast and swap out databases, web layers, and external services without rewriting everything. The big takeaway is to treat Flask, Django, ORMs, and cloud SDKs as adapters at the edges while your domain and use case code stays plain and boring.
“Mikhail Korolev: How to Hack the Job Market & Get Into VIP Rooms” from ConTejas Code
Podcast ⸱ Feb 16, 2026 ⸱ 01h 30m 58s
tldl: A software engineer shares how he went from getting dropped by Toptal to landing an engineering role by showing up where hiring happens: conference side events, invite only dinners, and small groups. Hints: use conference directories to find events, follow the Pac Man rule for mingling so you do not trap people in conversations, and treat CFPs like a game with short abstracts and reusable talk outlines. He also argues DevRel is a bad bet at companies with unfinished products and admits to using ChatGPT to speed through a LeetCode screen, saying it’s like working the system that already rewards game playing.