Latest / MBA Training AI / Model Context Protocol: what MCP actually fixed

Transcript

Model Context Protocol: what MCP actually fixed · MBA Training AI

0:00
Host: Welcome to the Leaders Insights podcast.
0:07
Host: Today's episode, the quiet handoff that changed what AI agents can actually do.
0:12
Host: In December 2025, Anthropic gave away infrastructure it could have charged a fortune for.
0:18
Host: Nine months later, was that generosity or was that strategy?
0:22
Expert: It was strategy dressed as generosity, and it worked.
0:26
Expert: The model context protocol, MCP, basically a universal adapter that lets an AI agent
0:33
Expert: talk to your tools, your files, your databases without custom glue code for each one, became
0:38
Expert: an open standard.
0:40
Expert: Anthropic didn't lose control by giving it away.
0:43
Expert: They set the terms everyone else now builds on.
0:46
Host: That sounds like the story every vendor tells about, open.
0:50
Host: Why should I believe it changed anything real?
0:53
Expert: As before MCP, connecting an agent to say your company's sales force and your internal
0:58
Expert: wiki meant two bespoke integrations that broke every time an interface changed.
1:04
Expert: Engineers called it the end times M problem.
1:07
Expert: Every model times every tool, a maintenance nightmare.
1:10
Expert: MCP collapses that into one protocol, one socket, many plugs.
1:15
Host: Fine.
1:16
Host: But everyone's screaming that 2026 is the year of autonomous agents doing your job.
1:21
Host: What's the consensus getting wrong?
1:23
Expert: The consensus thinks the bottleneck was intelligence.
1:26
Expert: It wasn't.
1:27
Expert: The models were smart enough two years ago.
1:30
Expert: The bottleneck was plumbing.
1:32
Expert: Agents couldn't reliably reach the systems where actual work lives.
1:35
Expert: MCP fixed the plumbing, not the brain.
1:39
Expert: And here's the uncomfortable part.
1:41
Expert: Better plumbing exposes how bad the brain still is at judgment.
1:44
Host: Meaning what, exactly?
1:46
Expert: Meaning an agent can now flawlessly connect to your billing system and confidently issue
1:51
Expert: the wrong refund to the wrong customer at scale.
1:55
Expert: Speed of access went up.
1:57
Expert: Quality of decisions did not.
1:58
Expert: O'Reilly's radar reporting through 2026 keeps landing on the same thing.
2:04
Expert: The failures aren't connection failures anymore.
2:06
Expert: They're reasoning failures that used to be hidden behind broken pipes.
2:10
Host: The labs would push back.
2:12
Host: OpenAI and Google both publish adoption numbers that look spectacular.
2:16
Expert: They do.
2:17
Expert: And cross-check every one of them.
2:19
Expert: But OpenAI says agent task completion jumped past 70% on their internal benchmarks.
2:25
Expert: Remember, they sell the agents.
2:28
Expert: That's a vendor grading its own homework.
2:30
Expert: Same with Google AI's figures on tool use accuracy.
2:34
Expert: Useful directionally.
2:35
Expert: Worthless as gospel.
2:37
Expert: MIT Technology Review ran independent evaluations this year and found real-world completion rates
2:43
Expert: on multi-step business tasks sitting far lower, closer to half.
2:48
Expert: Once you count the silent errors, nobody catches until an invoice is wrong.
2:52
Host: So companies rolling out agents right now are buying a faster way to make mistakes?
2:56
Expert: If they skip the boring part, yes.
2:59
Expert: The boring part is scoping.
3:01
Expert: The teams winning in 2026 aren't the ones giving agents the most autonomy.
3:06
Expert: They're the ones drawing the tightest fences.
3:09
Expert: Let the agent draft the refund, pull the data, format the report.
3:13
Expert: Keep a human on the trigger for anything irreversible.
3:16
Expert: It's a golden retriever with a company credit card.
3:19
Expert: Enthusiastic.
3:20
Expert: Capable.
3:21
Expert: Absolutely not allowed to check out alone.
3:24
Host: But isn't that just old-fashioned automation with a chatbot on top?
3:28
Host: Where's the actual shift?
3:30
Expert: The shift is composability.
3:32
Expert: The agent can now chain tools it was never explicitly programmed to combine.
3:36
Expert: Pre-MCP, you scripted every step.
3:39
Expert: Post-MCP, you described the goal.
3:42
Expert: And the agent assembles the path from whatever tools are exposed to it.
3:46
Expert: It's genuinely new.
3:48
Expert: It's also exactly why the fencing matters.
3:50
Expert: You gave up knowing the precise steps in advance.
3:53
Host: Give me a company that got this right.
3:55
Expert: Block, the payments company, formerly Square, built internal agents on MCP early and were
4:01
Expert: public about it.
4:03
Expert: Their move was to expose a limited, well-documented set of internal tools rather than everything
4:09
Expert: at once.
4:10
Expert: Narrow surface, clear permissions.
4:13
Expert: Their engineers reported fewer runaway behaviors, precisely because the agent couldn't reach
4:18
Expert: into systems it had no business touching.
4:21
Host: So the quiet handoff wasn't about capability.
4:24
Host: It was about who controls the standard.
4:26
Expert: Right.
4:27
Expert: Anthropic gave away the protocol to own the ecosystem's shape.
4:31
Expert: Smart.
4:32
Expert: But for anyone actually deploying, the lesson is smaller and more useful.
4:36
Host: Which is?
4:37
Expert: Before you connect a single agent this quarter, write down the list of actions it
4:41
Expert: may take without a human.
4:43
Expert: If that list includes anything you can't undo, cut it.
4:47
Expert: That one boundary saves you more than any model upgrade will.
4:50
Host: Sources for today's episode.
4:52
Host: The decoder, TechCrunch AI, Ars Technica AI, O'Reilly Radar, OpenAI, Vendor, AI Lab,
4:59
Host: KNuggets.
5:00
Host: That's your five.
5:02
Host: The full AI track, structure, no fluff, is at MBA-training.com.