The unit of knowledge-that is proposition, expressed linguistically in declarative sentences, the unit of knowledge-how is skill. The use of "knowledge" here refers to non-propositional uses like "know how to ride a bike", which are often passed over in the traditional position, which Ryle called intellectualism when he introduced the distinction between knowledge-that and knowledge-how in 1940s. A similar distinction between tacit and explicit knowledge was introduced by Polanyi in 1958, he was inspired by Merleau Ponty.
Intellectualism reduces skills to propositions, namely sets of instructions on how to do it, methodology. But intuitively, skill fuses structure and action, it is implausible that one can spin action out of structure alone. In Shaw's Man and Superman one of the characters says "those who can, do; those who can not, teach". What those who "do" have is something beyond methodology, a coach may be able to teach an athlete how to perform a stunt without having the skill to perform it himself. One defense is recourse to tacit knowledge-that, the coach knows but he is unable to explain. But there are other problems. Lewis Carroll described a tortoise which was given p, p → q, and the description of how modus ponens works, but still failed to grasp the inference of q from p. Skill is not reducible to instructions, in fact applying instructions properly is in itself a skill. Ryle developed a regress argument against intellectualism based on iterating this, similar to Wittgenstein's rule-following regress. See Löwenstein's Knowledge-how, Linguistic Intellectualism, and Ryle's Return for a critique of the more recent version of intellectualism defended by Stanley and Williamson.
An impressive slew of 20th century philosophers, in addition to Ryle and Wittgenstein - Heidegger, Merleau-Ponty, Sellars, Quine, Davidson, Brandom (Kant and Hegel can be named as precursors) - concluded, coming from very different backgrounds, that it is pragmatics that underlies semantics, i.e. knowledge-how is more fundamental than knowledge-that. Perhaps the most ambitious and developed is Brandom's reduction of knowledge-that to knowledge-how:
"While the meanings studied by semantics may not consist in the roles played by expressions in linguistic practice (meaning need not be identified with use), according to this view those roles must at least establish the connection between contents, meanings or semantic interpretants, on the one hand, and linguistic expressions on the other. The semantic pragmatist’s basic insight is that there is nothing apart from the use of expressions that could establish such connections".
Brandom's project in Making It Explicit is to present linguistic practice in non-semantic terms, and then reduce semantic categories to them, Szubka's On the Very Idea of Brandom’s Pragmatism is a very short summary of the project's motivation and challenges. Brandom takes advantage of Ryle's and Wittgenstein's insight that skill is not just an ability to act, it is an ability to perform a task, "up to standards", "according to rules". This does not imply some spelled out or even spellable standards, that would be like reducing skill to instructions, but communal judgement of a perfomance as adequate or not. "Knowing a rule is knowledge-how... performances come up to certain standards, or satisfy criteria", says Ryle, "there has to be a way to grasp a rule which is not an interpretation", says Wittgenstein. Language is a rule-governed, normative, activity. So semantics is to be reconstructed not from purely behaviorist or functionalist descriptions of practice, pragmatics must be normative. Specifically, he singles out one special type of linguistic practice as responsible for generation of semantics, where performances are assertions governed by (informal) rules of inference:
"The game of giving and asking for reasons is not just one game among others one can play with language. It is the game in virtue of the playing of which what one has qualifies as language (or thought) at all. I am here disagreeing with Wittgenstein, when he claims that language has no downtown... This is a kind of linguistic rationalism. Rationalism in this sense does not entail intellectualism, the doctrine that every implicit mastery of a propriety of practice is ultimately to be explained by appeal to a prior explicit grasp of a principle. It is entirely compatible with the sort of pragmatism that sees things the other way around".
The mechanics of Brandom's reduction of semantic content to inferential roles is complicated, and involves scorekeeping of commitments and entitlements by fellow speakers, see simplified version in his Articulating Reasons, and discussion in Reading Brandom. Of course, pragmatist proposals do not have to be as rationalist (Heidegger and Merleau-Ponty emphasize "embodied-embedded" aspect of meaning), or as structured as Brandom's (late Wittgenstein was sceptical of any linguistic generalities). One can also moderate by granting certain autonomy to knowledge-that in advanced stages of linguistic practice. Davidson can be interpreted along such lines, but he does not use the terminology.