Does your Flow Engineering framework make it trivial to build robust, high accuracy, high performance agentic flows? Can you do: const f1 = Flow.of(..) .retry(3) .catch(f2) .bestOf(3) .memo() .eval(f3, uiFlow);

The formula for production-ready, reliable agentic flows where 100% output accuracy matters. Step 1: Decomposition - break down the overall task into as many small processing nodes as possible (each executed by an LLM) and compose them. Step 2: Tuning - tune the prompt

Fruit of 3Rs in the west bay (10)

DevOps -> DevUseOps

Jasprit “Best in the World” Bumrah🔥👌

LXMs are tremendously useful and there is a lot of product innovation to do. But LXMs have hit a wall. (1) LLM scaling is plateauing. (2) LRM reasoning has a bit of Perpetual Motion Machine hype. You cannot build reliability from layers of unreliability.

Something off about "Vertical Agents" with specialized SaaS UI. Agents for Accounting. Agents for HR. Agents for Sales. But each with its own new SaaS UI trying to overlay or replace the underlying System of Record that has its own UI. Not everything will be a chat interface

Slack outage mid work week is the new equivalent of a power outage before a school exam. Experience great frustration and relief in equal measure.

Time to write an 'AI linter' for py. Bring static type checking, inference and completions even for complex functional data transformations. Could possibly be better than TypeScript!

In many domains, - AI has better skills than humans. - AI is also easier to communicate with than humans. The latter will lead to a lot more jobs lost to AI than the former.

Flow Caching made easy with flows DX.

Flow orchestration made easy with flows DX.

HTTP -> REST :: Flows (agents, multi-agents, MCP) -> FAST Internet age gave us HTTP with the ability to interact with diverse web nodes. But web applications needed REST apis with stateless representations and standardized resource formats. AI age gives us agents, multi-agents,

Junior devs: I vibe coded my entire app with AI. Senior devs: AI, help me identify possible root causes for this bug, then please step away from my keyboard. ps: with current AI code tools, 5 mins of unsupervised vibe coding = 5 months of technical debt.

AI+human products are best for 'iterative refinement'. Iterations need newer product features that are largely absent from current AI product UIs. WHITEBOARD INSTRUCTIONS - scribble quick drawings to better convey critical structure that is hard to do verbally DIFF VIEWS -

ai+human product patterns

Diagonal AI : it's like vertical AI but at a 45-degree angle

Turnaround time as a proxy measure for corruption.

MCP+UI is the new HTTP+HTML.

Investors: AGI! CEO: User talks, AI executes. CTO: User talks, AI auto builds an agentic workflow and then executes it. VP: User talks, AI auto builds an agentic workflow with human-in-the-loop and then executes it. Product: User talks, AI presents a visual editor to co build