All these startups will preach for customer-first and then put AI bot in their customer support.

documentation is the most underrated engineering activity it doesnt ship features it doesnt close deals but it carries context across time every hour you spend writing it clearly saves ten hours of someone guessing later bonus points since you can both have AI help generate docs based on code changes and feed then that documentation to AI chatbots etc.

The immediate future of ALL agents is coding agents, and although I could be wrong, I believe this surprising fact is going to be a HUGE boost for @typescript in particular. Let me explain. If your business wants an agent to assist with customer support, employee onboarding, outbound sales, or payroll, then the agent they need is actually a coding agent. The reason for this is quite simple: coding agents have an ability to leverage their training data to solve general-purpose problems, in ways shapes by the tools they have access to. An outbound sales agent assistant can talk to your knowledge base, pull some contacts from your CRM, analyze conversation history, do a web search to learn about each prospect, and then send email through your Gmail account to each prospect. Doing all of this stuff, and doing even more that the agent was never explicitly designed to do, requires the ability to write, test, and execute code for ad hoc, one-off problems. Only a coding agent can do that, and thanks to innovation at the level of the model and harness, a coding agent can do it well. Now, a true general-purpose coding agent can work in any code base, in any language, in any operating system, and with any tech stack. Of course, that type of coding agent is very useful to developers. However, it's overkill for most agentic systems. Most custom agents do not actually need to work with any code base or any language and on any operating system. They just need the ability to write code in some language (which has a lot of libraries) and execute on some platform. What is the ideal language and platform? I'd argue that @typescript fits the bill PERFECTLY. Since TypeScript compiles to Javascript, it can run securely, in a completely sandboxed way, inside V8 isolates, WASM, etc., all of which creates a compelling story for secure, efficient, and scalable custom agent execution. Moreover, because TypeScript adds types to Javascript, those types can be used to catch a lot of common bugs and runtime errors that a Javascript coding agent would have trouble catching in advance--allowing for far faster and more efficient solution of general-purpose problems. So, while general-purpose coding agents will of course need to support all programming languages, platforms, and tech stacks, custom agents are likely to be specialized -- while they will be coding agents, they don't need to work with any programming language, platform, or tech stack. They just need to work with one, and currently, the best option appears to be @TypeScript, for reasons of security, portability, type-safety, and efficiency. Is it any wonder TypeScript is home to some of the most amazing innovations currently happening in AI?

.@GammaApp hit $100M ARR with a team of 50. Profitably. 50M users, 600K paying subscribers. And they did most of it with zero sales team and zero marketing spend. Still ... CEO @thisisgrantlee wishes he'd added a sales team earlier. Grant broke down how at SaaStrAI 2026: 1⃣Build word of mouth before you spend a dollar on marketing. It's the one channel that amplifies everything else and can't be bought. 2⃣The first 30 seconds of your product have to be magical. Not good. Magical enough that users tell their friends unprompted. 3⃣Become a creator yourself before you run creator marketing. You can't coach a motion you've never done. 4⃣Dogfood to build conviction, not just catch bugs. Gamma ran two products in parallel for 6 months and let the energy decide which one to kill. 5⃣Pricing and packaging is never done. Seat vs usage vs API is a live question forever now. 6⃣Don't wait too long to add sales. This was Grant's biggest regret. They rode inbound so hard they never engaged their own self-serve base, and dropped real team and department demand sitting right in front of them. 👉Product-led doesn't mean sales-never. youtu.be/1i-CW4eknro?si=fSCF…

2026 is the year voice agents finally became good

Building memory for AI agents is less about storage and more about retrieval. Let me explain... I'm building a personal assistant from scratch for my next book with Manning. And one challenge I faced was deciding how the agent retrieves (from unified memory) the appropriate information quickly, reliably, and with as little complexity as possible. This led me to store the entire knowledge graph in a single database (e.g., MongoDB). One system handles: Full-text search Semantic search Graph traversal But you'll lose a graph-native query language. In exchange for simplicity, this is a trade I'd make every time for a personal assistant. When doing GraphRAG, the agent has three ways to search its unified memory: 1/ Graph search This is the default path. Run text search and vector search in parallel. Fuse both rankings using Reciprocal Rank Fusion (RRF). Traverse 2-3 hops through the graph to retrieve connected knowledge. Rerank the candidates. To keep context tight, pick the top 10. 2/ Deep search Sometimes keeping the top 10 isn't enough. You want to use everything that was retrieved as context. But after 2-3 hops, you can easily retrieve 50+ documents that won't fit in the context window. Instead, save everything to disk and build a lightweight LLM wiki on demand. This temporary wiki serves as the agent's localized memory, allowing it to explore large amounts of data through progressive disclosure without overwhelming the context window. 3/ Agentic search Some questions don't fit predefined retrieval algorithms so the LLM writes the database query itself. The ontology tells it which entities and relationships exist. You add a validation loop to ensure the query is syntactically correct. And a permission layer to ensure the query stays within safe boundaries. Here's the insight: GraphRAG isn't just vector search plus a graph; It's multi-hop traversal during retrieval. Similarity finds the entry point and the graph finds everything connected to it. We serve this unified memory as an MCP server via @fastmcp (by @PrefectIO). The agent never talks directly to the database. Instead, a harness such as Claude Code calls search-and-write tools, while the MCP layer decides how memory should be queried. Every fact the agent retrieves must first be processed. This is where @PrefectIO comes in. Every ingestion pipeline runs asynchronously as durable workflows across parallel workers with retries, caching, checkpointing, and centralized rate limiting. Even if one extraction step fails, indexing can continue, so the graph remains searchable. The read path never waits for orchestration. This has become one of my most important design principles. Orchestrate the writes. Never the reads. P.S. What retrieval strategy do you rely on most when building the memory for your AI agents?

this is one AI tool that understands your entire company. Tool of The Day - Day 21 name - Dust → what it does @DustHQ lets companies build AI agents that can securely connect to internal tools like Slack, Notion, Google Drive, GitHub, CRMs, and other business apps. instead of using AI as a standalone chatbot, teams create specialized agents that understand company knowledge and can help with real work across departments. → why people care most AI tools only know what you tell them in a single chat. Dust gives AI access to the knowledge your company already has, so answers are more accurate, workflows become automated, and teams spend less time searching for information or repeating the same tasks. → how people actually use it - build AI assistants for customer support, sales, marketing, engineering, HR, and operations - search company knowledge across Slack, Notion, GitHub, Google Drive, and other connected apps - generate reports, meeting briefs, documentation, and internal summaries - automate repetitive workflows with custom AI agents - share agents across teams so everyone benefits from the same knowledge and best practices → extra proof - serves more than 3,000 organizations - led by @gabhubert and @spolu - Dust has raised over $61 million in funding, - powers over 300,000 AI agents across customer teams.

TENCENT JUST DROPPED A NEW AI MODEL AND I TESTED IT IN 5 MINUTES ⬇️ I asked it to build an assistant that reads customer support messages and automatically decides: → what the issue is about → whether it's urgent or not → what response to give the customer I gave it a real case: "I've been locked out of my account for 3 days and I already paid this month's subscription." and in under 13 seconds it gave me everything back perfectly organized and ready to use. what surprised me the most: it's free to try right now, and it performs just as well (or better) than models that cost several times more. if you work with AI or automation for your business, this is one of the best things I've tested in terms of quality-to-price ratio. 🔖 save this to try it when you have 5 minutes free. #Hy3 #Hunyuan #AIAgents @TencentHunyuan

AI slop is not a tool problem. It's a taste problem. The machine gave everyone infinite paint, and most people immediately chose beige.

i have this file to replace the claude code system prompt with a pi-style one alias claude-pi='claude --system-prompt-file ~/.claude/pi-system-prompt.txt --dangerously-skip-permissions'

TENCENT JUST DROPPED A NEW AI MODEL AND I TESTED IT IN 5 MINUTES ⬇️ I asked it to build an assistant that reads customer support messages and automatically decides: → what the issue is about → whether it's urgent or not → what response to give the customer I gave it a real case: "I've been locked out of my account for 3 days and I already paid this month's subscription." and in under 13 seconds it gave me everything back perfectly organized and ready to use. what surprised me the most: it's free to try right now, and it performs just as well (or better) than models that cost several times more. if you work with AI or automation for your business, this is one of the best things I've tested in terms of quality-to-price ratio. 🔖 save this to try it when you have 5 minutes free. #Hy3 #Hunyuan #AIAgents @TencentHunyuan

Agentic AI adoption is on fire at @Uber, and it's changing the way we build, not just in engineering, but across the entire company. Today, 99% of our engineers use AI tools. More than 70% of pull requests are attributed to local or cloud agents. And our engineers have built 2,500+ agent skills across the software development lifecycle. Those numbers are exciting, but they led us to a much bigger question: How do we bring agentic AI beyond engineering? Finance. Legal. Operations. Marketing. Customer Support. HR. Procurement. These functions run on complex workflows that are often manual, highly nuanced, and spread across dozens of systems. You can't automate them effectively by looking at process diagrams or documentation. You have to understand how the work actually gets done. So we created something called Agentic Pods. The idea is simple. We handpicked ~30 of our most AI-proficient engineers (people with deep knowledge of Uber's systems) and paired each of them with a domain expert from a business function. Then we gave every pod just two weeks. • Days 1 – 2: Shadow the expert. Observe every step. Document workflows. Ask questions. Build intuition. • Day 3: Prioritize opportunities based on scale, repetition, business impact, and data availability. • Days 4 – 5: Build a working agent alongside the person doing the job. • Days 6 – 9: Validate with several others performing the same work. Does it generalize? Does it actually make their job better? • Day 10: Ship. In just the past two months, we've run 16 Agentic Pods across 16 different business functions. • Capital allocation across 150 cities: 15 hours → 30 minutes. • Financial pacing reports: 2 days → 10 minutes. • Marketing web quality assurance: 2 weeks → 50 minutes. • Support workflow creation: 9,000 manual workflows → self-service automation. The productivity gains are impressive, but what surprised us most wasn't the speed. • It was how quickly engineers embedded in unfamiliar domains uncovered opportunities that had been hiding in plain sight. • The biggest wins rarely come from automating one task. They come from rethinking an entire workflow. Once you redesign the workflow around AI, you often eliminate handoffs, remove unnecessary approvals, replace legacy tooling, reduce vendor spend, and dramatically accelerate decision-making. • The workflow becomes the unit of automation - not the individual task. • The most impactful agent skills cut across teams, orgs, functions, tools, and systems. The biggest lesson? The best AI opportunities are rarely visible from the outside. You discover them by sitting next to the people doing the work, understanding every friction point, and building with them, not for them. We're now forming a dedicated team to scale this further and go deeper. They'll deeply understand the work, redesign it from the ground up, and use AI to fundamentally change how the business operates. It's exciting times!

The tech workforce is splitting in two A year ago, we ran the first large-scale survey of how tech workers feel about their jobs and careers. What emerged we summarized in four words: burned out, but optimistic. Today, we're back with the results from our 2026 survey, and it's a tale of two workforces. Half of tech right now feels amplified by AI—more capable, more confident, more excited than they've been in their entire career. The other half feels shaken by it—less sure of their value and whether there’s still a place for them. Which side of that line you fall on predicts how you feel about your career more than anything else, including your role, seniority, company size, or any other measure we collected. The workforce is bifurcating into two realities. A few other takeaways that surprised us: + Significant burnout jumped from 44.7% to 55.7% in one year, while career optimism fell from 54.8% to 48.7%. A worrisome trend. + 53% of tech workers would steer a newcomer away from a career in their own role, even when they're optimistic about their own future. + The biggest AI fear is of being squeezed to do more work. Only 22% worry about “losing my job to AI.” Far more worry about being expected to do more for the same pay, getting trapped in an unsustainable pace, and the quality of their work declining. The question that best predicts how a tech worker feels about their work, in 2026, is no longer "What do you do?" or "Where do you work?" It’s "What has AI done to your sense of who you are?" Read the full report here: lennysnewsletter.com/p/how-t…

When I'm working on refactoring a large and relatively boring codebase, AI lets me go very very fast; I can usually run multiple agents at once very productively. When I'm working on something like the formal verification infrastructure I'm building right now, it's remarkable how much of the time I'm still blocked on my own thinking or doing specification review or on talking to experts about the problems I'm encountering. I'm working vastly faster than I ever could before, but it's not as productive as I'd like. Hopefully future AI agents will be capable enough that I don't have to make so many of the low level design decisions.

Founders handling DMs, solo builders replying to users, early-stage teams managing everything manually — everyone eventually hits the same wall: “Can you share the details again?” That one sentence kills trust faster than people realize.