Union.ai self-service deployment is now available on AWS Marketplace.

Learn more

The Union.ai Blog

Most AI pipelines fail quietly and get rerun by hand. Here is how teams build durable AI workflows and agents on Union and Flyte, from engineering deep dives and tutorials to what customers learned in production.

Start here
AI agents
Durable AI workflows
Flyte
× Clear
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

All posts

Develop and Debug Locally Before You Deploy

Develop and Debug Locally Before You Deploy

Run Flyte pipelines locally with flyte run --local for the same caching and error handling as a deployed run, then drop the flag to run the identical code on a cluster.

Sage Elliott
Sage Elliott
August 25, 2026
Union.ai
Flyte
AI
A Modern Alternative to Slurm for ML Workloads

A Modern Alternative to Slurm for ML Workloads

What one of computing's most successful schedulers got right, why ML teams are leaving it anyway, and what your workflows look like on the other side.

Samhita Alla
Samhita Alla
August 21, 2026
Union.ai
Flyte
Compare
Anatomy of a Durable Run

Anatomy of a Durable Run

A tour of the four durability mechanisms in Flyte 2 — checkpoints, crash recovery, cross-run recovery, and caching — with the code and design decisions behind each.

Niels Bantilan
Niels Bantilan
August 19, 2026
Flyte
AI
Agentic AI
Durability
Reuse Deployed Tasks Across Teams

Reuse Deployed Tasks Across Teams

Flyte remote tasks let teams call each other's deployed work by name, so an ML pipeline can use a data team's task without copying code, sharing a library, or building an API layer.

Sage Elliott
Sage Elliott
August 18, 2026
Union.ai
Flyte
AI
Pass Dataclasses, Pydantic Models, and Complex Types Between Tasks with Serialization

Pass Dataclasses, Pydantic Models, and Complex Types Between Tasks with Serialization

Flyte serializes dataclasses, Pydantic models, and other complex Python types automatically, so you can pass typed objects between tasks without writing JSON encoders.

Sage Elliott
Sage Elliott
August 11, 2026
Union.ai
Flyte
AI
Recover from Pipeline Failures with Retries and Self-Healing Tasks

Recover from Pipeline Failures with Retries and Self-Healing Tasks

ML pipelines fail. APIs time out. Spot instances get preempted. GPUs run out of memory — the question is how gracefully your pipeline recovers.

Sage Elliott
Sage Elliott
August 4, 2026
Union.ai
Flyte
AI
Flyte 2 Is Generally Available: The Durable, Open-Source AI Runtime

Flyte 2 Is Generally Available: The Durable, Open-Source AI Runtime

Durability across logic and infra. Reproducible, type safe, production scale.

Ketan Umare
Ketan Umare
August 4, 2026
Flyte
Union.ai
Data Processing
Training & Finetuning
Inference
Why Untrusted Kernel Evaluation Needs Process Isolation (and How We Built It)

Why Untrusted Kernel Evaluation Needs Process Isolation (and How We Built It)

Anyone who has evaluated LLM-generated GPU kernels at scale knows the pain: one kernel crashes, corrupts CUDA state, and suddenly nothing after it is trustworthy. Your kernel isn't broken — the one before it was.

Chloe Chong
Chloe Chong
August 3, 2026
Union.ai
AI
Training & Finetuning
Compare
A Memory Store Built on Flyte and Cognee

A Memory Store Built on Flyte and Cognee

Most "AI memory" today is a thin illusion. You stuff the last few turns of a conversation into the prompt, maybe sprinkle in some retrieved chunks from a vector store, and call it memory.

Muhammad Adil Fayyaz
Muhammad Adil Fayyaz
July 30, 2026
Union.ai
Flyte
AI
Agentic AI
Data Processing
Building Grounded Agents on Fresh Web Data

Building Grounded Agents on Fresh Web Data

Most "agentic AI" demos are not actually useful. They're impressive the first time you see them. They summarize things, answer questions, sound confident. T

Niels Bantilan
Niels Bantilan
July 30, 2026
Union.ai
AI
Financial Services & Fintech
Agentic AI
Partner
Stop Burning GPU Hours: Right-Size your AI Tasks with Flyte Resources

Stop Burning GPU Hours: Right-Size your AI Tasks with Flyte Resources

Not every task in your AI pipeline needs an A100. Your data preprocessing might run fine on 1 CPU and 512MB of RAM, while your model training task needs 8 GPUs and 64GB of memory.

Sage Elliott
Sage Elliott
July 29, 2026
Union.ai
Flyte
AI
From DNA to 3D Fold: Compare a Gene Across Six Species with Carbon and ESMFold

From DNA to 3D Fold: Compare a Gene Across Six Species with Carbon and ESMFold

In this post we'll take a single gene, pull it from six species that diverged up to 450 million years ago, score the DNA with a language model, translate it to protein, fold it into 3D structures, and build a phylogenetic tree.

Sage Elliott
Sage Elliott
July 28, 2026
Biotech & Healthcare
Data Processing
Union.ai

Nothing matches those filters. Try another topic or a broader search.

Clear all filters