Abhinav Asthana

Agents are about to consume APIs at 1000x the rate of today's apps — across runtime AND the whole SDLC: CI machines, workloads, debugging, incident response, MCP servers. The way we hand out API access wasn't built for that world. 🧵

The full breakdown from our team: blog.postman.com/ai-ready-apis-…

See where your API stands: postman.com/ai/ai-ready-ap…

API programs are undergoing one of their biggest changes they've ever faced: a completely new class of users. For decades, that user was a developer who read your docs, signed up, and emailed support when something was unclear. Today, agents can do that work. An agent can read your spec, wires up the integration, and ships, often without anyone visiting your site. And when it gets stuck, it doesn't open a ticket. It quietly moves to a competitor it can use, and you never find out. Most companies are already deploying AI agents, but only a quarter have designed their APIs for them. The teams that close that gap first are the ones that will pull ahead. So we built AI-Ready APIs, a free assessment that scores how ready your public API surface is for the agents already integrating it. Point it at your domain and get a graded, evidence-backed report of exactly where an agent gets stuck. All built from our team's years of experience embedded with some of the largest API programs in the world mobilizing towards agent readiness. Want to fix your agent experience? Postman's API Governance solutions keeps every API spec agent-ready, enforcing standards both at design time and in CI. Postman Workspaces become a curated context layer for partners (and their agents) to discover, execute, and integrate your APIs. Fern's platform help you transform your API specs into agent-ready docs, SDKs, and CLIs. Start optimizing your product for agents today. Links in the thread...

Excited to see support for the latest updates to the MCP protocol in @getpostman.

@GergelyOrosz My observations on this situation: x.com/a85/status/208…

And this is exactly why “SaaS replaced by AI coded software” is not viable for any company that has a an ACTUAL business. It’s a fun exercise to rebuild ticketing/o11y/feature flags etc… but do you really want to maintain it? Do you want updates, oncall etc to be YOUR problem??

My unscientific analysis of token spend reveals that only ~20% of the tokens are involved in doing any useful work today.

Computer use is only cool if the service you're trying to use has a bad API

gave up on waiting on nikita for the articles fix - part 1 is here part 2 is coming x.com/i/article/2078…

@stevekrouse If I am not wrong, a gateway gateway is a federated gateway according to Gartner. :)

Introducing the AI Gateway Gateway Why be trapped into one AI Gateway when you can use them all? Please, nobody make a competing AI Gateway Gateway, because then someone else will have to make an AI Gateway Gateway Gateway and this shit will never end

🚀We’re Raising the Bar for Enterprise AI Agents. Today, we’re releasing APIFlow-Bench. The first benchmark that aims to test how AI agents can execute real enterprise workflows end-to-end. In production, “almost works” is not acceptable: → One bad assumption early → cascading rework → One unconfirmed write → broken automation → One expired token → compliance exposure Even if the agent somehow does accomplish the task, the astronomical token costs for agentic workflows has been proof that fail, fail, and try again is not an acceptable solution. Most benchmarks still measure fluency for answers from LLMs, APIFlow-Bench measures survival. ⚫️ What is APIFlow-Bench? A deterministically validated, sandboxed benchmark for long-horizon API workflows that mirrors how agents actually fail in the enterprise. It evaluates what matters for production agents: • Multi-turn dependency & workflow fidelity. • State carryover & context preservation. • Recovery from failure & safe retry behavior. • Behavior under enterprise constraints (auth, quotas, error codes). ⚫️ Why This Matters Because in a world where agents are wired directly into APIs, queues, and orchestration layers, the blast radius of a small error is no longer small. A minor hallucination at step 3 becomes a production incident at step 27. A model that can survive long API chains is not merely better at tool use. It is better at preserving context, respecting constraints, managing state, and completing work across dependent steps. That’s why API readiness is such a strong proxy for agent readiness. Check out the blog post describing the benchmark, the methodology and results here: blog.postman.com/apiflow-bench The public APIFlow-Bench leaderboard is at: postman.com/ai/apiflow-lea… with full transcripts here: github.com/postmanlabs/ap… In our technical blog, we break down: • How we designed the capability axes. • How we generated and validated realistic API workflows. • Why we use deterministic validators instead of LLM judges. • What the early results tell us about agent reliability in production. If you’re building, buying, or betting on AI agents for the enterprise, this is the benchmark you probably want to check. Check out the post to see the winners and stay tuned for the next update with the latest OpenAI, Anthropic, Open-weight (Kimi, Qwen) models! #AI #AIAgents #EnterpriseAI #APIs #Benchmarks #LLM #Automation #DevTools #Postman

AI is making building faster, but few have noticed how in some companies AI is significantly reducing corporate BS: useless meetings, illogical debates, docs no one reads, performance theater, revisionist history, etc. And eliminating this BS makes work much more fun.

@zeeg We are building this in @astropods_ which is framework agnostic. astropods.com

IMO going forward all agent frameworks should be disregarded unless they bake-in solutions for: 1) end-to-end testing that also tolerates llm judge assertions 2) local and production telemetry, both traces and unhandled errors I find myself constantly fighting with both of these concerns in new technology. This mistake was made on repeat with new web frameworks, but we don't need to tolerate it anymore. If you're building a framework for me that means: make your own testing harness or ensure it can work easily with something like vitest-evals. Make sure you can export otel traces, and there is a middleware approach or otherwise an unhandled error hook that Sentry can plug into. Eve does the former _really_ well if you want an example. Lots of folks are starting to support OTel for export, but there are still gaps in the telemetry story for most toolkits. Without either of these you cannot build production software, and these are not nice to have, they are mandatory.

@adityaag We’d have probably turned on Fable if it was willing to do the work. Saw many times just in personal use that it would degrade to Opus.

I am literally switching models off of Fable right now for our systems. This isn't trying to be hyperbolic....but why would you pay the price if there is a good and free alternative?

Great article. This is why we are building @astropods_ and running all our agents on it.

Is it normal for Claude to ask for access to tools before it starts doing work? Seeing this even if the task has no need for access.

Writing with AI is ironically the worst use case of generative AI. Glad to see the sentiment being shared by others.

Check out details here and sign up for early access: postman.com/product/fabric…

Postman’s Fabric Gateway Early Access is now open! Postman Fabric has been designed from the ground-up for agents with support for MCPs, CLIs, as well as APIs - both AI and deterministic APIs. We have moved all our AI and MCP workloads to Fabric already and it’s been a game changer to have one control plane. Not only does it simplify administrative burden of managing all these APIs, it's also a much better experience for developers and agents.

Dear fella designers: APIs first, UI last. And keep it dumb!