I built a score for AI-agent readiness — now I’m testing whether companies actually care
I’ve been building MachineReach, a public-hostname scanner for AI-agent readiness.
The question behind it is simple:
If an AI agent lands on a website, can it actually discover what the service does and how to interact with it?
MachineReach checks public evidence such as OpenAPI, MCP, A2A Agent Cards, ARD/ai-catalog, agent plugins, documentation, trust signals, and other machine-readable interfaces.
I spent a while calibrating the scoring model against real websites, then froze Score Model v1.0.
Now I’m deliberately stopping feature work and testing whether companies actually care enough to act.
My current experiment:
20 targeted API/agent companies → 10 open their report → 3 attempt a recommended fix → 1 is willing to pay for verification + monitoring
Some patterns from the scans so far:
ordinary human-readable website: around 40
one validated machine interface: around the 70s
multiple validated agent/API protocols: 80+
MachineReach doesn’t just show a score. It tells you exactly what is missing, simulates the expected score after each fix, rescans to verify the improvement, and can monitor the hostname for regressions.
What I’m trying to learn now:
If you run an API, developer tool, or AI product, would a report like this actually make you change something — or is it only interesting information?
I’d genuinely value criticism from people building APIs, MCP servers, agents, or developer infrastructure.
This is exactly the question I’m trying to validate now.
MachineReach already shows the exact fix path and simulates the score change, but I’ve intentionally stopped adding features until I know whether teams actually care enough to act on the gap.
I like your Impact/Action idea though — especially “who is affected” and “which fix should come first.” If the current experiment shows people understand the technical gap but still don’t act, that may be the missing layer.
Right now I’m tracking: report opened → remediation viewed → fix attempted → score improved → monitoring → willingness to pay.
So hopefully the next 20 conversations tell me whether the bottleneck is understanding, urgency, implementation, or budget.
Exactly. That’s why I’m deliberately treating the 20 → 10 → 3 → 1 experiment as more important than getting more traffic.
My guess is the hardest jump will be “interesting report” → “we’re actually going to change our machine interface because of it.”
If companies make the fix but don’t want ongoing verification/monitoring, then the scanner has value but the recurring business model may be wrong.
I’ll share the funnel numbers once I have enough real outreach data.
That’s a useful test. I’ll be interested to see what the funnel looks like once you have enough outreach data to separate those two stages.
Absolutely — that’s the part I’m most curious about too. I’ll share the numbers once the sample is meaningful enough.
Thanks. I’d be happy to continue the conversation privately as you get more data. What’s the best email to reach you on?
sure, prismgridai@gmail.con