Cineasta de mentiras · Escritor frustrado · Software engineer

Joined November 2008
"La libertad es como la felicidad: nunca se llega. Nunca se tiene completa." —Pedro J. Gutiérrez (Animal tropical)
1
3
Kevin Vicent retweeted
anatomy of a managed deep agent. langch.in/mda
33
196
12
1,716
337,025
Kevin Vicent retweeted
You have to invent the future. It doesn't just happen.
257
783
102
8,316
239,418
📌
tienen algún hábito que les haya mejorado mucho la salud mental?
1
26
Kevin Vicent retweeted
how do you trust your agents' work more? copy this post/link and give it to your @bot or agent! 1. codebase improvements: this is just good software engineering - better architecture, stronger types and invariants, clear boundaries. agents love to copy what exists and take the shortest path, so make the easy thing the right thing. categorically eliminate problems through code and you won't need to think about it again. 2. static analysis: lint rules, tests, CI, verification (eg TLA+). very token efficient ways to catch issues ahead of time. 3. prose: comments, rules, AGENTS.md, skills - anything that can't be enforced and requires agent judgment. use it minimally, but do use /create-verification-skill. 4. human review: the hardest to scale. at the rate agents land code, it's pretty much impossible to thoroughly review every PR, so spend that effort on the highest leverage point, designing an environment where agents do the right thing by default. this forms an inverse pyramid: you want most of your effort at the top, least at the bottom. codebase improvements are deterministic, making it more token efficient, and lets you sleep more soundly at night. install github.com/cursor/plugins/tr… and create a new Project: 1. look through my slack/chat messages and code review comments where i've given feedback on PRs, code, architecture 2. use /recall on my past transcripts to learn how i steer agents 3. /poteto-mode use /correct and /architect on past commits to find the most common pitfalls agents fall into, then propose a plan to rearchitect the code i own so it's more agent friendly. answer open questions with prototyping rather than defer to me, and back the plan with real data. 4. before you begin, use /sharpen + /in-your-own-words and confirm this prompt with me
603
514
110
6,876
734,380
Kevin Vicent retweeted
ojo con cambiar el modelo en medio de una sesión cuando trabajan con agentes como Codex o Claude Code. imaginen que llevan 2 horas trabajando con Opus y deciden pasar a Sonnet para resolver algo más sencillo. Sonnet no puede reutilizar la caché de Opus, así que tiene que volver a procesar el contexto que recibe. en sesiones largas, esto puede aumentar bastante los costos y la latencia. ¿qué pueden hacer? una alternativa es delegar esa tarea a un subagente que utilice Sonnet. de esta forma, mantienen Opus como modelo principal, mientras Sonnet trabaja con un contexto independiente que tiene solamente la información necesaria. cuando termina, devuelve el resultado y siguen trabajando con Opus. así pueden aprovechar modelos más económicos para tareas puntuales, sin cambiar el modelo principal ni pasarle todo el historial de la conversación. la clave está en darle a cada modelo únicamente el contexto que necesita para resolver su tarea.
45
31
6
788
62,990
Kevin Vicent retweeted
"There's no silver bullet that results in 90% efficiency gains" ...But this article is a great primer on how to approach AI Transformation!
1
3
40
7,140
Kevin Vicent retweeted
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source! Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable. 🔗 github.com/deepseek-ai/DeepI… 1/4
1,588
4,834
2,309
42,672
6,526,671
Kevin Vicent retweeted
DeepSeek just open sourced its high-performance, large-scale inference framework DeepInfer in Apache-2.0 license. github.com/deepseek-ai/DeepI… Supported features: - Optimized kernels, model and sampler runtime - Parallelism: Attention DP / balanced CP, MoE EP + EPLB - DSpark and confidence-scheduled verification - SWA bounded replay - Asynchronous scheduling (zero-overhead batch scheduler) - Batch-invariant inference - Deterministic sampling - Structured output
211
208
148
2,588
296,350
This is a very strong pattern: Decision models + ontologies
Don't waste all your time improving how your AI agents reason. Do this instead... Spend more time defining what they should reason about. Even the smartest model can make a bad decision if the options are poorly defined. This is why I like using Jev and ontologies together... Jev by @typesafeai turns unstructured inputs into structured decisions with calibrated confidence scores. Instead of asking an LLM to reason through every decision, you define the possible outcomes in code. Jev selects an outcome and returns its confidence. And the ontology defines what those outcomes mean. Imagine an AI agent handling customer support tickets. The ontology defines: • Which issues belong to billing versus technical support. • How customers, subscriptions, and invoices relate. • Which categories should never overlap. Jev classifies tickets against those categories. Based on its confidence, your agent can: 1. High confidence → Execute an action. 2. Medium confidence → Escalate to an LLM. 3. Low confidence → Request human review. So expensive reasoning is reserved for cases that need it. But what if your ontology is wrong? Jev can confidently choose the wrong category if the options are poorly defined. You still need evals to identify failures, improve definitions, and test changes before deploying. As those definitions evolve, version them like your prompts, datasets, and code. The takeaway: Building better AI agents isn't always about making the model smarter. Sometimes it's about defining its world more clearly. P.S. How are you defining the decision space for your AI agents?
16
Kevin Vicent retweeted
Don't waste all your time improving how your AI agents reason. Do this instead... Spend more time defining what they should reason about. Even the smartest model can make a bad decision if the options are poorly defined. This is why I like using Jev and ontologies together... Jev by @typesafeai turns unstructured inputs into structured decisions with calibrated confidence scores. Instead of asking an LLM to reason through every decision, you define the possible outcomes in code. Jev selects an outcome and returns its confidence. And the ontology defines what those outcomes mean. Imagine an AI agent handling customer support tickets. The ontology defines: • Which issues belong to billing versus technical support. • How customers, subscriptions, and invoices relate. • Which categories should never overlap. Jev classifies tickets against those categories. Based on its confidence, your agent can: 1. High confidence → Execute an action. 2. Medium confidence → Escalate to an LLM. 3. Low confidence → Request human review. So expensive reasoning is reserved for cases that need it. But what if your ontology is wrong? Jev can confidently choose the wrong category if the options are poorly defined. You still need evals to identify failures, improve definitions, and test changes before deploying. As those definitions evolve, version them like your prompts, datasets, and code. The takeaway: Building better AI agents isn't always about making the model smarter. Sometimes it's about defining its world more clearly. P.S. How are you defining the decision space for your AI agents?
17
28
4
140
4,907
Kevin Vicent retweeted
Generative UI is close to seeing its moment emerge – and I can't wait to show you all a new CF experience. Cloudflare's recent models make rendering near instant. Clef = Choose the best component Clairic = Populate props from image/text Rendering 5-10 components in ~1 second. Creating an amazing user experience has a lot of layers too it. It's not just about speed. Output needs to be exact. Styling needs to be consistent. Components need to be effortlessly composable. And none of that happens without an amazing context layer to glue it all together. It's more than just the models. And all of this can be complementary to any LLM output – not dependent upon it. Really excited with where things are at internally and when we can show more of what we're building to the world.
75
66
15
1,471
71,031
Kevin Vicent retweeted
The biggest AI workflows, ripe for agent-ification in each department, that we've seen at Varick: 1. Finance a. Manage AP exceptions: most AP teams are managing 600-800 exceptions a month where invoices don't match POs or receipts and require manual resolution b. Automate your month end close: background execution of bank recs, accruals, intercompany, etc. can get your close cycle from 18-22 days to 7-9 days c. Cash application and collections: automate matching remittance advice to open invoices and getting customers to pay on time 2. Sales Ops a. Manage CRM data and rep admin: we see sales reps spend 38% of their time updating Salesforce b. Deal desk approvals: we see deals take 23 days to get discounts, pricing, etc. approved c. Prepare for forecast calls: ensure all the right data is available for each deal to prepare for Monday forecast calls (call notes, stage changes, next steps, etc.) 3. HR a. Manage payroll changes: compare your ATS, HRIS, and payroll systems before payroll to make sure all new hires, raises, terminations, etc. are accurate. Most payroll corrections are due to issues before the payroll run. b. Handle employee inquiries: we've seen up to 80% of inquiries for PTO, benefits, policy questions, etc. can be handled without human intervention c. Onboarding: manage the process from offer acceptance to day 1, including background checks, equipment, account provisioning, I-9, etc. 4. Legal a. Redline contracts: reduce turnaround time from 5+ days to same day for first pass redlines on NDAs, MSAs, vendor contracts, etc. based on your contract playbook b. Contract intake: manage the intake process for contracts, including determining the requestor, type of contract, contract template, contract lawyer, etc. c. Track contract obligations: extract contract renewal dates, notice periods, auto-renewal provisions, etc. from signed contracts 5. Operations a. Disruption management: move from a week-long process to manage shipment delays, supplier issues, etc. to a few hours b. Order management exceptions: many orders have exceptions that require manual management, whether it's a customer on credit hold, bad ship address, lack of inventory, etc. 6. Procurement a. Supplier onboarding: reduce the time it takes to onboard new suppliers from 26 days to 4 by automating collection of W-9s, banking info, insurance, sanctions, etc. b. 3-way match: compare invoices to POs and receiving documents to identify instances of duplicate invoices, invoices over PO price, etc. 7. IT a. Joiners, movers, and leavers: ensure employees have all the right access on day 1, instead of day 4, and remove access the day they leave b. Ticket routing: we see 1 in 3 IT tickets are routed to the wrong team and require rerouting 8. Marketing a. Speed to lead: enrich, score, and route inbound demo requests to the right sales rep in under 5 minutes b. Media reconciliation: reconcile media invoices to insertion orders and media actually run to identify the 8-12% of media spend that is lost 9. Customer Operations a. Route escalations to the right team the first time, instead of having them bounce back and forth between support and engineering multiple times b. Create case summaries to ensure your engineering team doesn't send cases back for additional details All of these can be run in the background using your existing systems (NetSuite, Salesforce, SAP, ServiceNow, Dynamics, etc.) and exceptions can be routed to your people. If you can solve these for your clients, you'll make them very happy. But in order to succeed across the entire company, you have to go about this systematically. Read more on how to go about it below:
49
40
3
440
32,461
Kevin Vicent retweeted
Using state machines for agent logic is extremely slept-on Don't think of them as static workflows. They're like LEGO bricks that the agents can rearrange + improve on autonomously, in a human-readable way Otherwise it's like shaping markdown Play-Doh and hoping it holds
this has been working for like a month and I've had to restart it maybe twice
37
39
6
781
35,507
A few weeks ago, we built a new harness, AgentRun, which turns repetitive steps into workflows and code using Jev. This week we're open-sourcing its best features as @pidotdev extensions! Day 1: Pi-vision uses Jev to give your agents eyes without wasting frontier tokens. Analyzing videos, images, or documents (a common use case @grepdotai) with frontier models is extremely token-inefficient. With this extension, @typesafeai's Jev does the analysis, while your Pi agent does the orchestration. The result is a faster and cheaper workflow for detection and analysis at scale! You can use this workflow for document verification, understanding image content, and needle-in-a-haystack search in videos.
8
8
3
68
4,667
Kevin Vicent retweeted
"it's underrated how hard it is to build agents!" great convo w/ @hwchase17 and @HarryStebbings on @twentyminutevc an agent = model + context + harness, and you can improve an agent by tweaking any of the three. a few takeaways: 1. the highest roi thing you can do is start by optimizing context (then the harness, then the weights); this is where domain knowledge lives and that's incredibly valuable 2. the story of continual learning is: run an agent, collect traces, aggregate learnings, fix something, measure the fix, loop! the fix can land in any of the three layers. those fixes all look different but the cycle is the same. 3. memory is still massively unsolved, but also massively important. the hardest part is distilling learnings from traces which is incredibly domain specific. 4. when you build agents, you're probably worried about a few things: accuracy, cost, latency, and security. we've seen these themes confirmed year after year in our dev surveys 5. when we think about reducing cost, the pipeline looks like adding visibility into spend, putting controls in place, and only then pushing on agent optimizations. this is how we reduced our coding agent spend @LangChain 6. model routing, which is great for cost controls, is most effective in the harness (where the context needed for a routing decision lives) 7. open models are used for more narrow tasks right now (small parts of the agent loop OR for very specific domains w/ post-training), but as they improve they'll power more core agent loops 8. we're looking at a future with both horizontal agents and vertical agents. horizontal agents will use the vertical agents, and infra needs to change to support this load (agents can do background work at a rate that humans cannot). youtube.com/watch?v=1jDCPEHw…
22
13
1
119
20,904
Kevin Vicent retweeted
one thing that made this click for me — with decision models, you have to define categories and it returns probabilities that the input matches those categories. but so much of the data we interact with is completely new to us. how can you define categories for something you've never read? with clairic, you can extract values directly from the inputs, without any premeditated categories. the pipeline becomes clairic to extract values → clef to make decisions with those values → LLMs to take actions on those decisions. all of these models run on Workers AI.
23
65
13
897
40,147
Kevin Vicent retweeted
Introducing Clairic, from Cloudflare. A System One extraction modeI, capable of extracting, computing, estimating, and forecasting data from unstructured inputs. cfl.re/4zIgfpH
53
201
103
1,661
163,958
i had opus 5.5 distill this 30 min lecture about pi-durable into a 7 min vid with my voice. > qwen3 tts could do a better job cloning my voice > the video looks great. i'm working on industrializing these kind of explanation videos. good starting point, but i still recommend watching the full lecture for a more detailed explainer.
i just recorded a full tutorial on pi durable 1:48 Checkpoints + crash recovery 8:38 Harness vs. app 13:46 App architecture + storage 19:30 Six setup steps 23:11 Local Qwen + llama.cpp demo + restart recovery
16
39
5
519
57,327
Kevin Vicent retweeted
For people who keep asking what to build as a Backend Engineer in the AI Era. ➣ Build your own AI Gateway (semantic caching + token budgeting + provider fallback chains) ➣ Build your own Durable Workflow Engine (state checkpointing + saga patterns + human-in-the-loop pauses) ➣ Build your own PII Redaction Proxy (UUID masking + deterministic encryption + post-generation unmasking) ➣ Build your own Hybrid Search Engine (pgvector + BM25 + cross-encoder reranking + row-level security) ➣ Build your own Streaming SSE Proxy (token chunking + backpressure handling + TTFT/ITL metric emission) ➣ Build your own Token-Aware Rate Limiter (Redis sliding window + prompt complexity scoring + per-tenant quotas) ➣ Build your own Webhook Reconciliation Engine (idempotency keys + dead-letter queues + automated replay tooling) ➣ Build your own Event-Sourced Trajectory Store (immutable agent logs + time-travel debugging + state replay) ➣ Build your own Inter-Agent Auth Broker (delegated OAuth flows + scoped JWTs + least-privilege tool access) ➣ Build your own Cost Kill-Switch Middleware (spend projection + circuit breakers + graceful degradation chains) ➣ Build your own Shadow Traffic Router (5% async mirroring + trajectory diffing + silent cost comparison) ➣ Build your own Queue-Based GPU Autoscaler (queue depth tracking + spot instance provisioning + drain logic) ➣ Build your own LLM Tracing Pipeline (OpenTelemetry spans + token counting + correlated DB query logs) ➣ Build your own Edge Fallback API (cloud 503 detection + local model routing + offline request queuing) ➣ Build your own Ephemeral AI Provisioner (Terraform scripts + isolated vector DBs + per-PR test environments) Pick 3. Build them from scratch. Document every decision. Most people write CRUD APIs. Builders architect the nervous system for autonomous agents. Bookmark this. You'll need it.
22
75
4
579
24,681