If it was me in charge, I'd be making the case to stay in Afghanistan as long as necessary --- it's the central front in our ongoing GWOT (with AFPAK). It's been a long time, though, and with Barack Hussein in power, we have no leadership in foreign affairs.
In any case, check out the useful analysis from Stephen Biddle, Fotini Christia, and J Alexander Thier, at Foreign Affairs, "Defining Success in Afghanistan: What Can the United States Accept?"
We won't build a centralized democracy in Afghanistan, but a decentralized model might work:
Power sharing would be easier under a decentralized democracy, in which many responsibilities now held by Kabul would be delegated to the periphery. Some of these powers would surely include the authority to draft and enact budgets, to use traditional alternatives to centralized justice systems for some offenses, to elect or approve important officials who are now appointed by Kabul, and perhaps to collect local revenue and enforce local regulation.
Increasing local autonomy would make it easier to win over Afghans who distrust distant Kabul and would take advantage of a preexisting base of legitimacy and identity at the local level. The responsibility for foreign policy and internal security, however, would remain with the central government, which would prevent even the more autonomous territories from hosting international terrorist groups or supporting insurrection against the state.
A decentralized democracy along these lines should be an acceptable option for the United States. Its reliance on democracy and transparency is consistent with American values. Individual territories with the freedom to reflect local preferences may adopt social policies that many in the United States would see as regressive. But the opposite could also occur, with some places implementing more moderate laws than those favored by a conservative center. By promoting local acceptance of the central government, this option would remove much of the casus belli for the insurgency. And it would preserve a central state with the power and incentive to deny the use of Afghan soil for destabilizing Pakistan or planning attacks against the United States.
A decentralized democracy would comport with much of the post-Cold War experience with state building elsewhere. A range of postconflict states in Africa (Ethiopia and Sierra Leone), Europe, (Bosnia and Macedonia), the Middle East (Iraq and Lebanon), and Asia (East Timor and, tentatively, Nepal) have used some combination of consociationalism, federalism, and other forms of decentralized democratic power sharing. Although it is too early to make definitive claims of success, to date not one of these states has collapsed, relapsed into civil war, or hosted terrorists. And some, such as Bosnia and Ethiopia, have remained tolerably stable for over a decade. This is, of course, no guarantee that decentralized democracy would work in Afghanistan. But its track record elsewhere and its better fit with the country's natural distribution of power suggests that it offers a reasonable chance of balancing interests and adjudicating disputes in Afghanistan, too ....
Afghanistan is not ungovernable. There are feasible options for acceptable end states that would meet core U.S. security interests and place the country on a path toward tolerable stability. The United States will have to step back from its ambitious but unrealistic project to create a strong, centralized Afghan state. If it does, then a range of power-sharing models could balance the needs of Afghanistan's internal factions and constituencies in ways that today's design cannot, while ensuring that Afghanistan does not again become a base for terrorists. In war, as in so many other things, the perfect can be the enemy of the good. The perfect is probably not achievable in Afghanistan -- but the acceptable can still be salvaged.