I am a crank WRT the semantic web. Sure, there will be pockets of success, but I think the whole concept is very 19th century. It is almost the opposite of the network effect; it is not of much use unless a significant portion of the sites on the web are marked up. Its whole search mechanism is logic based, while much out there in non-logical. There is all that W3C navel gazing to learn. Just about anything designed by a committee, in anticipation of a revolution, fails.

I think the semantic web will fail because more revolutionary technologies will pop up out of nowhere and surprise us all.

Alternatively, the pockets of success can be significant. I have zero hope of it being "search engine" worthy, since it's too easy to fake (see meta keyword).

But... in more controlled environments, it's a very powerful tool. Corporate networks and Medical Research are two areas where it's getting bigger, and has made significant advances.

I love the idea, but it's aimed wrong every time it's explained. Trust is a big issue, and there's no way to prevent lies. But in the absence of trust problems, it's incredibly powerful.

"semantic web" is a buzzword, too ill-defined to allow anyone to define themselves with it label of "semantic web programmer."

The intellectual academics point at the rise of standards, structured information and shout "the semantic web is here!"

The pragmatic implementers point at the reams and reams and reams of still unstructured data and wonder where the academics point.

Save yourself the trouble; don't worry about what the semantic web is or isn't. Don't worry about how to leverage it. Find a problem and just solve it.

Data-based startups will be implement internally before it hits popularity. This is supported by semantic-esque elements have shown up in many APIs (such as Amazon).

In my startup, we're using ARC (http://arc.semsol.org/) and microformats internally; however, we're not doing it just for the sake of new technology - data is managed better, databases can be designed well, and there is a performance gain in some cases.

Upmodded. My firm produces semantic analysis software for Chinese text:

http://popupchinese.com/words/downloads

The major issue hampering adoption is that unlike services based on pattern-matching, services based on semantic/NLP analysis don't have a simple mechanism to hide errors.

There is probably a threshold at which error rates are tolerable for consumer-facing services, but we are not there yet. So the uses are largely on the backend in areas with narrowly-defined problems where semantic tech can help with automation and content processing and where people have an incentive to improve performance over time by customizing the software itself. This is happening, but you don't see glowing write-ups in Wired magazine since the uses are extremely field specific and usually not terribly sexy.