← Extractions

Forms Are Dead: Building Agentic Workflows in Ruby — Adam Okoń at wroclove.rb 2026

Adam Okoń presents a pattern for augmenting complex B2B forms with LLM-driven analysis — users paste requirements as text and the agent translates them into pre-filled form state for human review. Covers monolith vs. microservice architectures with Ruby LLM, async/fibers for heavy IO workloads, and observability via OpenTelemetry, with a production case study showing dramatic time reduction and higher adoption of supporting features.

Model

claude-opus-4-7

Ingestion

f15b33e0

Input tokens

434,528

fresh: 126,418
cached: 130,066
cache write: 178,044

Output tokens

9,183

Duration

163.2s

Roundtrips

Tool calls

Cost

$0.00

Nodes/edges extracted

17 / 31

Read set (nodes/edges)

444 / 2

Nodes (17)

update Adam Okoń person

attrs (empty) → {"role" => "independent consultant / tech lead / product developer", "community_role" => "co-organizer of local Ruby ...

description Ruby developer building agentic AI workflows. → Independent consultant working mostly as a tech lead / product developer building B2B solutions. Ruby developer of 17...

short_description Ruby developer, agentic workflows. → Independent Ruby consultant, tech lead building B2B solutions, agentic workflows.

update Forms Are Dead: Building Agentic Workflows in Ruby talk

attrs {"type" => "talk"} → {"date" => "2026-04-17", "type" => "talk"}

description Talk on agentic AI workflows. → Adam Okoń's first big-stage talk, delivered at wroclove.rb 2026. Argues that complex forms are a bottleneck for non-t...

short_description Talk on agentic AI workflows. → Adam Okoń's wroclove.rb 2026 talk on augmenting B2B forms with LLM-driven agentic workflows in Ruby.

create Augmented Form Workflow concept

kind (empty) → concept

name (empty) → Augmented Form Workflow

slug (empty) → augmented-form-workflow

attrs (empty) → {"category" => "pattern"}

description (empty) → Pattern introduced by Adam Okoń for complex B2B forms that are primarily used to translate requirements (e.g. email f...

short_description (empty) → Two-step form pattern: paste requirements as text, LLM analyzes and pre-fills the form for user review.

create Consolidation Window concept

kind (empty) → concept

name (empty) → Consolidation Window

slug (empty) → consolidation-window

attrs (empty) → {"category" => "pattern"}

description (empty) → Insight from Adam Okoń's talk: while waiting for LLM tokens during the analysis phase, the same background job can pr...

short_description (empty) → Using the LLM analysis phase to prepare state for subsequent screens, amortizing cost across features.

create Baseline Metrics Before Building AI Features takeaway

kind (empty) → takeaway

name (empty) → Baseline Metrics Before Building AI Features

slug (empty) → baseline-metrics-before-building-ai-features

attrs (empty) → {"type" => "recommendation"}

description (empty) → Before building an AI-augmented workflow, record baseline metrics (especially time of the operation) over some period...

short_description (empty) → Measure operation time and token cost baselines before and after to decide if an AI feature has impact.

create Use async for Heavy IO LLM Workloads takeaway

kind (empty) → takeaway

name (empty) → Use async for Heavy IO LLM Workloads

slug (empty) → use-async-for-heavy-io-llm-workloads

attrs (empty) → {"type" => "recommendation"}

description (empty) → For LLM-heavy, IO-bound background work in Ruby, prefer fibers over threads. The async community provides an adapter ...

short_description (empty) → Switch Active Job adapter to async and use async's barrier to scale concurrent LLM IO calls.

create Best UI Is No UI takeaway

kind (empty) → takeaway

name (empty) → Best UI Is No UI

slug (empty) → best-ui-is-no-ui

attrs (empty) → {"type" => "insight"}

description (empty) → Adam Okoń's personal opinion delivered in Q&A when asked why he chose an AI approach rather than splitting the comple...

short_description (empty) → Adam Okoń's opinion that the best user interface is the absence of one.

create Community Distinguishing Heavy CPU vs Heavy IO takeaway

kind (empty) → takeaway

name (empty) → Community Distinguishing Heavy CPU vs Heavy IO

slug (empty) → community-distinguishing-heavy-cpu-vs-heavy-io

attrs (empty) → {"type" => "insight"}

description (empty) → Observation from Adam Okoń: across wroclove.rb 2026 talks and recent Ruby tooling, the community is increasingly dist...

short_description (empty) → Ruby community increasingly separates heavy-CPU and heavy-IO concurrency patterns.

update Ruby LLM tool

description Ruby gem providing a cleaner, more readable wrapper over LLM APIs than early LangChain-style scanning for magic marke... → Ruby gem providing a cleaner, more readable wrapper over LLM APIs than early LangChain-style scanning for magic marke...

update async tool

description Newer Ruby concurrency framework based on fibers. Not available when David Halasz started the project; he notes he wo... → Newer Ruby concurrency framework based on fibers. Not available when David Halasz started his VNC proxy project; he n...

update Puma tool

description Concurrent Rack web server used by David Halasz to run the PURR demo server. → Concurrent Rack web server used by David Halasz to run the PURR demo server. At wroclove.rb 2026 Adam Okoń noted that...

update Anthropic company

description AI company that coined the term 'many-shot jailbreaking' and published the corresponding paper on LLM jailbreaking te... → AI lab that publishes frontier LLMs and coined the term 'many-shot jailbreaking'. Referenced in Adam Okoń's wroclove....

short_description AI safety company that published the many-shot jailbreaking paper. → AI safety company and LLM lab; publisher of Claude and the many-shot jailbreaking paper.

create Why AI backend instead of optimizing the UI into smaller steps question

kind (empty) → question

name (empty) → Why AI backend instead of optimizing the UI into smaller steps

slug (empty) → why-ai-backend-instead-of-optimizing-the-ui-into-smaller-steps

attrs (empty) → {"answer_summary" => "Personal opinion: the best UI is no UI."}

description (empty) → Audience member referenced an earlier talk showing a complex form split into multi-step pages each containing only a ...

short_description (empty) → Why not split the 20-field form into multi-step pages of 2–5 fields instead of adding AI?

create Why server-side AI instead of in-browser WebLLM question

kind (empty) → question

name (empty) → Why server-side AI instead of in-browser WebLLM

slug (empty) → why-server-side-ai-instead-of-in-browser-webllm

attrs (empty) → {"answer_summary" => "In regulated markets the security department will likely ban client-side AI; server-side keeps ...

description (empty) → Audience member suggested running the LLM in the browser (WebLLM, future embedded Chrome models) to prefill forms dir...

short_description (empty) → Why run AI on the backend when modern browsers (and WebLLM, future embedded Chrome LLM) can do it client-side?

create Measured async performance vs sequential and threads question

kind (empty) → question

name (empty) → Measured async performance vs sequential and threads

slug (empty) → measured-async-performance-vs-sequential-and-threads

attrs (empty) → {"answer_summary" => "Sequential ~50s → async ~15–20s; async's abstraction is easy to extend and scale further."}

description (empty) → Audience question on whether Adam could quantify the gain from using fibers and async over a traditional thread-based...

short_description (empty) → Did switching from sequential/thread-based calls to async+fibers show measurable IO gains?

create Preserving user options when AI-prefills a constrained form question

kind (empty) → question

name (empty) → Preserving user options when AI-prefills a constrained form

slug (empty) → preserving-user-options-when-ai-prefills-a-constrained-form

attrs (empty) → {"answer_summary" => "Keep the form — AI only pre-fills state; users still see/pick options manually, and you need a ...

description (empty) → Audience question: complex forms communicate to the user the structure and the options available for each field; how ...

short_description (empty) → Forms give users the structure/options they can pick — how do you preserve that with AI?

create Walking users through nested conditional forms with AI question

kind (empty) → question

name (empty) → Walking users through nested conditional forms with AI

slug (empty) → walking-users-through-nested-conditional-forms-with-ai

attrs (empty) → {"answer_summary" => "First audit/simplify nested paths; then show the user a live preview of the LLM's understanding...

description (empty) → Audience question about complex forms where selecting one option scopes subsequent options, and how to help users con...

short_description (empty) → How do you guide users through deeply nested conditional forms so they know the AI understands them correctly?

Edges (31)

update Adam Okoń → authored → Forms Are Dead: Building Agentic Workflows in Ruby

context (empty) → Adam Okoń delivered this talk as his first big-stage presentation.

update Forms Are Dead: Building Agentic Workflows in Ruby → presented_at → wroclove.rb 2026

context (empty) → Delivered on day 1 of wroclove.rb 2026.

create Adam Okoń → has_skill → Ruby

attrs (empty) → {"level" => "expert", "years" => "17+"}

context (empty) → Uses Ruby after 17 years and still uses it mostly as his tech stack.

relation (empty) → has_skill

source_node_id (empty) → 668173b1-698e-4806-852f-79ba9803b350

target_node_id (empty) → c7f25b33-06c9-460e-aca7-fe993123ebee

create Forms Are Dead: Building Agentic Workflows in Ruby → about → Augmented Form Workflow

context (empty) → Core pattern introduced in the talk.

relation (empty) → about