Co-Founder / CEO @diagridio Dapr co-founder https://xitter.cf/t.co/YNRbFJtJtE and Ex-Microsoft
USA
Joined August 2008
- Tweets955
- Following111
- Followers1.8K
- Likes643
Just end my day at AGNTCon/MCPCon here in San Jose. Spoke to so many people and had an amazing time taking with a ton of developers building agents and needing durable execution as essential for long running solutions. And most said "No SaaS, I want BYOC or full disconnected"
I also walked the floor and counted the sponsor lineup.
~16 of ~110 sponsors are MCP/AI gateways: Solo, Kong, Gravitee, Traefik, WSO2, TrueFoundry, Runlayer, Obot, MintMCP, Stacklok, Airia, Cequence. That’s the most crowded single category. That one will consolidate fast. They are generally useful, but in the long run are a temporary security solution, in the light of agent identity being the required solution.
The rest of the map:
🔐 Agent identity & security, ~14: Auth0, Keycard, WorkOS, Descope, Aembit, Chainguard
🔌 MCP connectors & tooling, ~12: Workato, CData, Boomi, Merge, Apify, Speakeasy
📊 Observability, ~9: Datadog, Elastic, Chronosphere, groundcover
💸 FinOps & monetization, ~7: Finout, PointFive, Stigg, Revenium
🤖 Agent frameworks & durable execution, ~12: AWS, Strands, Google ADK, Mastra, Temporal, Orkes, Diagrid
The gateway layer is crowded. Which other categories consolidate?
MCP reminds me of WSDL in 2001.
WSDL gave us a standard way to describe a service: operations, messages, where to send them. Feed it to a tool and you got a working client. Integration got cheap, and then everyone asked the next questions. Who may call this? What happens on a retry? What happens if the caller crashes halfway through?
Web services answered with WS-Security, WS-ReliableMessaging, WS-AtomicTransaction and a long tail after them. Few teams could implement the stack and fewer could debug it. REST won by doing less and moving the hard problems into the runtime.
MCP looks like it learned that lesson. The July spec took things out: no more sessions, Tasks moved to an opt-in extension, and Roots, Sampling and Logging were deprecated. It’s a thinner protocol, built to sit behind ordinary infrastructure.
That still leaves the hard questions with every team running an MCP server. Which tools can each caller use? Where do the upstream credentials live? And if an agent crashes after calling a refund tool, does it refund again?
Those answers belong in the runtime, not the protocol: platform identity, credentials injected at request time, retry policy, and every tool call recorded as a durable step.
I wrote about why MCP should stay thin and where those answers belong instead:
medium.com/@mark.fussell/mcp…
Mark Fussell retweeted
Goose can already do the work. Now it survives the crash.
Goose agents run inside Dapr Workflows. Crash mid-task, restart, and completed turns aren’t re-run.
diagrid.io/blog/durable-exec…
Your agent shouldn’t decide what happens when it fails.
Neither should another LLM.
Erlang worked this out on phone switches decades ago. What agents can take from supervisor trees, and the one thing Erlang never needed:
lnkd.in/p/gMt8wggR
Your AI agent crashes at step nine. What happens next?
I compared 17 agent frameworks: LangGraph, CrewAI, PydanticAI, Google ADK, AWS Strands, the OpenAI and Claude agent SDKs, Mastra and more.
They differ a lot on graphs, loops and roles. On a crash, most behave the same way. They either restart the run, or resume from the last checkpoint and repeat whatever was in flight, including the LLM call you already paid for and the tool call that already sent the email.
The fix is durable execution: record every LLM and tool call as it completes, and resume exactly where it stopped. Some now call this the agent harness.
It doesn’t have to come from your framework. Diagrid Catalyst provides it underneath any of them, so you keep the framework you chose.
Full comparison and the eight criteria I used: diagrid.io/blog/ai-agent-fra…
What happens when your agent crashes halfway through?
I just published Orchestration vs Choreography Was Never a War. Interested in your thoughts 🤔medium.com/@mark.fussell/orc…
Check out my latest article: No Workflow Is an Island. No Agent Is Either. linkedin.com/pulse/workflow-…
Mark Fussell retweeted
We made a game about crashing. In Datacenter Run you steer the Diagrid logo through a datacenter full of chaos. It will crash. Dapr replays it from history and you keep going.
Keyboard only, plays in the browser: diagrid.io/arcade/datacenter
#Dapr #DurableExecution
This annual event is happening tomorrow, with an amazing set of speakers 👇
Dapr Day 2026 is happening tomorrow! Join this free CNCF community event if you want to build reliable and durable applications. Learn directly from the maintainers and end users.
Agenda:
- The Dapr co-creators on where the project is heading in 2027
- Five production practices from engineers running Dapr today
- Durable Spring AI applications built on Dapr Workflows
- Cutting the local Dapr inner loop down to seconds
- Distributed event processing patterns
- A maintainer's view of Dapr Agents
Free and virtual, Thursday Oct 8, 15:00 to 18:30 UTC.
Register at community2.cncf.io/events/de…
Mark Fussell retweeted
Dapr Day 2026 is happening on October 8! A free CNCF community event with talks from Dapr maintainers and end users.
On the agenda:
- Dapr in 2027 from the Dapr co-creators
- Durable Spring AI with Dapr Workflows
- A local inner loop in seconds
- 5 production-ready Dapr practices
- Distributed event processing
- Dapr Agents from a maintainer perspective
Register now: community2.cncf.io/events/de…
I just published "Checkpointing Is a Comfort Blanket"👇
medium.com/@mark.fussell/che…
Mark Fussell retweeted
Join the Dapr Community Call on Sep 30th!
Topics:
- Building a Reliable Distributed Job Execution Engine with Dapr, EF Core, and the Transactional Outbox Pattern
- What's new in the .NET and JS SDKs
The best place to watch is our YouTube channel: youtube.com/live/oSYNlkEUGsM…
Mark Fussell retweeted
New episode of The AI Kubernetes Show Mark Fussell with William Morgan on what the cloud-native era already learned about reliability, and what the agent era is relearning the hard way.
hubs.ly/Q04x4YY_0
#Kubernetes #AgenticAI #Dapr @mfussell
Mark Fussell retweeted
"Do developers really get to that point and just go, 'I guess we've got to start over'?"
"Literally, yes."
Mark Fussell on CIO's DEMO series, killing the infrastructure under ten banking agents mid-transaction and watching them resume — then showing how Catalyst decides which agent is allowed to reach which MCP tool.
Same app run twice: plain LangGraph, then with Catalyst underneath. 15 minutes.
Read it here:
hubs.ly/Q04wGXhx0
Or watch it on youtube: hubs.ly/Q04wGXkj0
#AgenticAI #MCP #DurableExecution @mfussell
Precisely why you need tamper-proof attestation for agents to prove what they have done. diagrid.io/blog/what-is-agen…
Mark Fussell retweeted
Debugging Dapr applications locally means inspecting multiple processes, monitoring log outputs, component files, and state stores.
The new free, open-source Diagrid Dev Dashboard gives you a live view of everything Dapr on your local machine:
- See apps, sidecars, components, actors, subscriptions, and more
- Inspect workflow executions and event history
- Terminate or purge stuck workflow instances
- Build components and resiliency policies in a UI
- Send test messages to verify subscriptions
It's a single binary with zero setup. Try it now: diagrid.io/dev-dashboard
Mark Fussell retweeted
Time from vulnerability disclosure to first exploitation:
2021: ~1 year
2026: ~1 day
2027 (projected): ~1 minute
Mean time to remediate: ~95 days.
The EU Cyber Resilience Act is landing in the middle of that gap.
Read the latest blog by Mark Fussell, Diagrid's CEO and Co-Founder — scope check, what the CRA expects of the Dapr inside your product, and an 11-point readiness checklist: hubs.ly/Q04svk6L0
#CyberResilienceAct #OpenSource #Dapr #SupplyChainSecurity #Compliance
Excited to announce Agentic Durable Execution is available in Diagrid Catalyst today. You can read more about how this solves the needs of agents in my post below👇and in this NewStack post thenewstack.io/diagrid-catal…
Durable execution is now Agentic: introducing Diagrid Catalyst 2.0.
Now teams can make their existing agent workflows resilient to failure and cryptographically provable, without re-architecting how those agents are built. Agents automatically resume from the exact point of failure, and every step can be cryptographically signed and traced back to its source.
It brings durable and verifiable execution to the AI frameworks developers already use: LangGraph, Microsoft Agent Framework, Google Agent Development Kit (ADK), AWS Strands, OpenAI Agents SDK, CrewAI, and more.
Most of those frameworks have no built-in way to guarantee a long-running workflow recovers and finishes when something breaks. Work is lost, tokens burn, and someone has to step in.
Keep the framework you build on. Add a Diagrid code package, gain durable workflows immediately.
Read the release 👇
hubs.ly/Q04q_2Pb0
#AIAgents #DurableExecution #PlatformEngineering #LLMOps #DiagridCatalyst
