Back on Track

Now that Twitter has gone public with its Summize acquisition and Evan Williams’ detailed discussion with Mike Arrington at Foo Camp, we can put to rest the garbage that Twitter is not perhaps the most important service of the next generation of computing. Williams’ transparency about potential business models may be of most interest to the TechCrunch audience, but for those who see the world through an enterprise lense, the Summize deal marks the end of speculation and the start of real work in harnessing the real-time cloud.

Conversational search, as John Borthwick describes it, has basic requirements for scalability and stability. The service has to manage spikes of non-real time data around events, software releases, political issues, and other rapidly swarming loads. Twitter has in recent days shown the ability to manage such peaks, though the overall flow does not yet match the peak of Twittermania that lead to the collapse and reworking on the basic systems.

Adding the one-to-one or one-to-many attributes of a real-time conversation in that environment has additional requirements. First of all, the response time of a discoverable conversation must be under 10 seconds, whereas asynchronous swarming can adjust (as it has had to since Track over IM was lost in May) to latency approaching 10 minutes as in the case of the Twitterspy hack. Note that Twitterspy’s delay is not a technical limitation but rather a business relationship limitation.

Simply put, for someone, anyone, to signal another person or group over the Twitter network, the roundtrip for a conversation to occur must be close to if not the same as the interval between when the signal is received and the signaler sees the response. If I post a Twitter message and track via Summize on my username (stevegillmor) the average response time is somewhere around 10 seconds. When Track over Gtalk was enabled, the delta was somewhat smaller, perhaps 5 seconds. As long as the signaler and I use keywords as part of each message or the signaler is following, we can chat back and forth as though this were a nailed-up IM channel.

Once conversation is possible, the next requirement is filtering, to reduce or eliminate cross-talk (spam) and flow affinity conversations into appropriate containers. Spam can be inadvertent (a Reporters bot that gathers and to some extent promotes certain voices but pollutes track conversations with repetitive duplication) or malevolent, the repeated violation of an implicit contract between people who largely are engaged in conversation based on respect and rules of discourse.

Summize’s filtering tools easily take care of most cases, adding the flag -Reporters to the search query. And keywords could also be used to encode conversations on the fly to scope conversations around topics or events of the moment to bring in more of what Scoble calls the noise of the cloud. Hashtags are one example of that, but there can also be more ad-hoc constructions once syntax is formalized for these processes.

Dave Winer’s experiments with URL shortening also expand the 140-character window for filtering, but new interfaces will need to be released to harness these tools for conversations. Speciifcally, XMPP or near-real-time API feeds will need to flow into URL-aware containers that can extract that data in a companion window or part of the screen. Since URLs are nothing less than triggers to execute code (HTML and/or Javascript in the example of a page or virtually any instruction over XML) clients will rapidly emerge with the “smarts” to orchestrate these signals and intermesh rich media. Seesmic and Twhirl already are doing this in a primitive way.

Beyond interface comes the oft-cited problem of how to consume this stream of data, the “you can’t just sit in front of your monitor all day” critique. Never mind that that is exactly what the financial markets require with ever more rapid flow of data, analysis, and response every minute of every day. With the iPhone and other smartphones now proliferating, who do we know who isn’t already in this game when they’re “away” from the office?

Twitter conversations harness the affinity group filtering of important data – with conversational tracking adding a dynamic ability to enter and leave information in real-time that is harvested by the group mind, not the assumptions of the individual about what is happening. This elasticity is clearly valuable – why do we camp on Mike Arrington’s doorstep if not for his insights and particularly those informed by his network of sensors in the disruptive technology community. This is a result of the success and now revenge of the RSS wave, where we have to find strategies to more efficiently cope with the firehose.

Just as RSS was a strategy to cope with the burgeoning Web of independent creators of value, Twitter and particularly Track media are the strategy to cope today. It’s not about sitting in front of your machine all day, it’s about cooperating and competing with other information-savvy groups in a common pursuit of value per minute of consumption. Add in the unfettered ability to weigh in to the stream – contouring and refining, scoping and amplifying, synthesizing and distilling. We will be rewarded for these skills individually and as partners.

In the ’90s the word for this was groupware – Lotus Notes, the Web, Office, IM. XML abstracted the differences between data formats and enabled a rollup of enterprise message buses and workflow processes in the major platforms. Virtualization at the hardware and OS levels led to the emergence of cloud computing and the baking of groupware into the cloud computing stack as a service, e.g. Salesforce, Google Apps, MobileMe. Recently, groupware has gone by a new name – social media – as we harvest our ability to author our social identities into filtered clouds of like interests and concerns.

The intersection of social media and the enterprise is littered with noise and panic as IT grapples with the implications for users and corporate interests. As Evan Williams suggests, preserving control of Twitter messages from the ravages of the unfettered Web free-for-all is a concern not just of business model but of user privacy. If Twitter moves too far in the direction of controlling the conversational real time flow, whether on the company’s behalf or as a proxy for corporate control, it will inevitably cause a user revolt or the rapid entry of bigger players.

Imagine the Twitter-enabled groupware of tomorrow. Your client manages multiple identities or filtered versions of your personal and professional selves, across Twitter and cooperating sub-clouds such as FriendFeed, Facebook and Gnip-hosted smaller sites, via IM channels including Gtalk, AIM, Yahoo, and Live Mesh. Conversational pings will register as alerts on the desktop, or via SMS, email, or IM on mobile devices, and archived for later scanning in cloud stores like Gmail’s chat archive.

Filter queues will allow smart clients to decorate the real time follow stream with visualization prompts for hot button threads, job-related announcements, time-sensitve appointment and event data, and social-graph aware behavior of affinity group. The more important threads are to people more important to you, the higher in the sort of priority they live, and the better chance of you seeing it in the time you have to react. Once engaged, who and for how long you interact, as well as the data the correspondents throw of in URL citations, amount of time talking, and the fan out across new nodes who track, follow, and engage – all of this becomes valuable data to refine and enhance the data stream on all participants’ clients.

Twitter has put to rest the notion that this discoverable conversation model is not valuable by investing a reported 10% of the company’s value in the Summize acquisition. As Williams observes, just because they are not yet ready to pursue revenue does not mean they don’t have at least a good idea of some business model(s), as well as some even better ones that may emerge. Summize took technology focused on a different problem space and business strategy and diverted it to the emerging Twitter data stream to great effect. And Twitter’s instability problems clearly stemmed from not fully grasping the virality of the service and particularly its most valuable uses. Track was a one-night coding add-on a year ago.

Essentially, the Summize acquisition is the result of a conversational swarm between the two companies and users. With the Summize CTO now head of combined engineering and the integration of fully one-third of the total engineering staff, Twitter has profited from a discoverable affinity conversation that continues to find fuel in the FriendFeed evolution as well as open source communities attracted by the opportunity to hold Twitter’s feet to the fire of a viable contract with both users and developers. If Williams and Arrington’s conversation and subsequent messaging about the deal proves a reliable indicator of the approach Twitter is taking, it will be good news for Track fans, the company and its investors, and the enterprise development community.