We recently confronted the thorny problem what to do when an executive agency, exercising delegated legislative authority, seeks to overrule a judicial precedent interpreting a congressional statute. In our constitutional history, after all, judicial declarations of what the law is haven’t often been thought subject to revision by the executive, let alone by an executive endowed with delegated legislative authority. Still, in recent years the Supreme Court has instructed us that, when a statute is ambiguous and an executive agency’s interpretation is reasonable, the agency may indeed exercise delegated legislative authority to overrule a judicial precedent in favor of- the agency’s preferred interpretation. See Chevron, U.S.A., Inc. v. Nat. Res. Def. Council, Inc., 467 U.S. 837, 104 S.Ct. 2778, 81 L.Ed.2d 694 (1984); Nat’l Cable & Telecomms. Ass’n v. Brand X Internet Servs. (Brand X), 545 U.S. 967, 125 S.Ct. 2688, 162 L.Ed.2d 820 (2005). And that development required us to confront this question: accepting that an agency may overrule a court, may it do so not only prospectively but also retroactively, applying its new rule to completed conduct that transpired at a time when the contrary judicial precedent appeared to control? De Niz Robles v. Lynch, 803 F.3d *11441165 (10th Cir. 2015). Now that curious question has returned, this time with a twist.
Our story starts with two provisions buried in our immigration laws: 8 U.S.C. §§ 1255(i)(2)(A) and 1182(a)(9)(C)(i)(I). The first statute “grants the Attorney General discretion to ‘adjust the status’ of those who have entered the country illegally and afford them lawful residency.” De Niz Robles, 803 F.3d at 1167. The second “provides that certain persons who have entered this country illegally more than once are categorically prohibited from winning lawful residency ... unless they first serve a ten-year waiting period outside our borders.” Id. Observers have long noted the tension between the two laws and wondered which should control. Employing the usual tools of statutory interpretation, this court in 2005 determined that the Attorney General’s discretion to afford relief without insisting on a decade-long waiting period remained intact. Padilla-Caldera v. Gonzales (Padilla-Caldera I), 426 F.3d 1294, 1299-1301 (10th Cir. 2005), amended and superseded on reh’g by 453 F.3d 1237, 1242-44 (10th Cir. 2006).
That judicial declaration of what the law is turned out to be anything but the last word. Not because the Supreme Court disagreed. But because in 2007 the Board of Immigration Appeals (BIA) issued In re Briones, 24 I. & N. Dec. 355 (BIA 2007). There the BIA offered its view that — -as a matter of policy discretion — the statutory tension should be resolved against affording the Attorney General any discretion to consider applications for adjustment of status when § 1182(a)(9)(c)(i)(I) applies. A conclusion directly at odds with the one we reached in Padillas-Caldera I. When the agency later sought to apply its new rule announced in Briones to a petitioner in this court, we agreed that the two statutory directives were ambiguous; that “step two” of Chevron required this court to assume that Congress had delegated legislative authority to the BIA to make a “reasonable” policy choice in the face of this statutory ambiguity; and that the Supreme Court’s extension of Chevron in Brand X further required this court to defer to the agency’s policy choice and overrule our own governing statutory interpretation in Padilla-Caldera I. See Padilla-Caldera v. Holder (Padilla-Caldera II), 637 F.3d 1140, 1148-52 (10th Cir. 2011).
But even that was hardly the end of it. Everyone accepts that, after Padilla-Caldera II, all future petitioners must satisfy the ten-year waiting period and may not seek discretionary relief from the Attorney General. But what about petitioners who applied for discretionary relief in express reliance on Padilla-Caldera I, before the BIA’s announcement of its contrary interpretation in Briones’1. In De Niz Robles, the BIA sought to apply Briones retroactively to foreclose any chance of discretionary relief for this class of persons. This court disallowed the attempt, holding that because the agency’s promulgation of a new rule of general applicability under Chevron step two and Brand X is an exercise of delegated legislative policymaking authority, it is subject to the presumption of prospectivity that attends true exercises of legislative authority. 803 F.3d at 1172-74.
The BIA isn’t one to give up, though. Today it brings us a new case that involves a (slight) variation. Like Mr. De Niz Robles, Hugo Gutierrez-Brizuela applied for adjustment of status in reliance on our decision in Padillar-Caldera I during the period it remained on the books. About that much there is no dispute. But unlike Mr. De Niz Robles, Mr. GutierrezBrizuela applied for relief during the period after the BIA’s announcement of its contrary interpretation in Briones yet be*1145fore Padilla-Caldera II declared Briones controlling and Padilla-Caldera I effectively overruled. The BIA suggests this factual distinction makes all the legal difference. But we fail to see how. Indeed, the government’s position in this appeal seems to us clearly inconsistent with both the rule and reasoning of De Niz Robles.
Take the rule first. De Niz Robles held that Briones was not legally effective in the Tenth Circuit until this court discharged its obligation under Chevron step two and Brand X to determine that the statutory provisions at issue were indeed ambiguous, that the BIA’s interpretation of them was indeed reasonable, and that Padillar-Caldera I was indeed overruled. As we explained, “[a]n agency in the Chevron step two/Brand X scenario may enforce its new policy judgment only with judicial approval. So, for example, the BIA depended on Padilla-Caldera II to render Briones effective.” Id. at 1174 n.7. Until this court handed down Padilla-Caldera II, then, Padilla-Caldera I remained on the books as binding precedent in the Tenth Circuit on which litigants were free (and expected) to rely, and Briones bore no legal force. Yet, despite De Niz Robles’s clear holding on this very score, the BIA today seeks to apply Briones to conduct in this circuit that predates Padilla-Caldera II — when Padilla-Caldera I was the controlling law of this circuit and Briones was not. That De Niz Robles expressly forbids. Cf. Bankers Trust N.Y. Corp. v. United States, 225 F.3d 1368, 1376 (Fed. Cir. 2000) (noting that where “the original decision was based on direct judicial construction of [a] statute, not deference” to the agency, it remains “the law of this circuit” until it is “overturned” or a “later amendment to the statute is effective”).
Next consider the reasoning. In De Niz Robles we explained that, to the extent the executive is permitted to exercise delegated legislative authority to overrule judicial decisions, logic suggests it should be bound by the same presumption of pros-pectivity that attends true legislative enactments. 803 F.3d at 1172. After all, agents usually depend upon (and are limited to) the powers enjoyed by their principals. See generally Restatement (Second) of Agency § 17 (1958). And we know that, if Congress had sought to amend the law to unseat a judicial decision like Padilla-Caldera I, absent some clear direction otherwise (and subject to constitutional limitations on retroactive legislation), its actions would have controlled conduct arising only after the legislation went into effect. See Landgraf v. USI Film Prod., 511 U.S. 244, 270-73, 114 S.Ct. 1483, 128 L.Ed.2d 229 (1994). And here Briones went into effect in this circuit only when this court handed down Padilla-Caldera II. Meaning that individuals like both Mr. De Niz Robles and Mr. Gutierrez-Brizuela would have been free to rely on Padilla-Caldera I. Neither, as we explained in De Niz Robles, can we think of a sound reason why persons should be left in worse shape simply because they are the subjects of delegated legislative action rather than subjects of true legislative action. Indeed, as we noted in De Niz Robles, the Supreme Court itself has expressly recognized that “[cjongressional enactments ... will not be construed to have retroactive effect unless their language requires this result. By the same principle, a statutory grant of legislative rulemaking authority will not, as a general matter, be understood to encompass the power to promulgate retroactive rules unless that power is conveyed by Congress in express terms.” 803 F.3d at 1172 (quoting Bowen v. Georgetown Univ. Hosp., 488 U.S. 204, 208, 109 S.Ct. 468, 102 L.Ed.2d 493 (1988) (internal quotation marks omitted)). And, of course, no one before us contends that Congress has expressly conveyed to the *1146BIA the power to declare its rules retroactive.1
The due process and equal protection concerns that animated our holding in De Niz Robles also apply to this case. In De Niz Robles we explained that legislation is presumptively prospective in its operation because the retroactive application of new penalties to past conduct that affected persons cannot now change denies them fair notice of the law and risks endowing a decisionmaker expressly influenced by ma-joritarian politics with the power to single out disfavored individuals for mistreatment. See 803 F.3d at 1169-70. These very same concerns would arise if the BIA could apply Briones retroactively to Mr. Gutierrez-Brizuela’s conduct. After all, back in 2009 the law expressly gave Mr. Gutierrez-Brizuela two options: he could seek an adjustment of status pursuant to Padilla-Caldera I or accept a ten-year waiting period outside the country. Relying on binding circuit precedent, he chose the former path. Yet the BIA now seeks to apply a new law to block that path at a time when it’s too late for Mr. Gutierrez-Brizuela to alter his conduct. Meaning that, if we allowed the BIA to apply Briones here, Mr. Gutierrez-Brizuela would lose the seven years he could’ve spent complying with the BIA’s ten year waiting period and instead have to start that waiting period now. The due process concerns are obvious: when Mr. Gutierrez-Brizuela made his choice, he had no notice of the law the BIA now seeks to apply. And the equal protection problems are obvious too: if the agency were free to change the law retroactively based on shifting political winds, it could use that power to punish politically disfavored groups or individuals for conduct they can no longer alter. See id. at 1168-69, 1174-75.
Beyond looking directly to the due process and equal protection considerations that underlie the presumption of legislative *1147prospectivity, in De Niz Robles we also consulted both the doctrinal rubric the Supreme Court suggested in SEC v. Chenery Corp. (Chenery II), 332 U.S. 194, 67 S.Ct. 1760, 91 L.Ed. 1995 (1947), for deciding when to give retroactive effect to agency action, and our own similar formulation in Stewart Capital Corp. v. Andrus, 701 F.2d 846 (10th Cir. 1983). We found these authorities and their factors also counseled against allowing the BIA to apply its decision to individuals like Mr. De Niz Robles. De Niz Robles, 803 F.3d at 1175-80. And we find they do the same thing today. Neither should any of this come as a surprise, for the factors discussed in Chenery II and Stewart Capital simply seek to provide a doctrinal rubric for assessing the underlying due process and equal protection implications associated with retroactive agency action. And whether we analyze those concerns directly (as we just have) or indirectly (using these doctrinal rubrics), the answer should of course be the same. Id. at 1180.
To decide whether to permit retroactive agency action, both Chenery II and Stewart Capital essentially counsel us to “weigh” the costs the petitioner would encounter against the benefits the agency would enjoy. See id. at 1177. And in this case, as in De Niz Robles, it seems to us that balance tips decidedly in one direction. Consider the costs we would impose on petitioners. Normally, people are entitled to rely on judicial precedents as definitive interpretations of what the law is so long as those precedents remain on the books. See U.S. Bancorp Mortg. Co. v. Bonner Mall P’ship, 513 U.S. 18, 26, 115 S.Ct. 386, 130 L.Ed.2d 233 (1994); AT&T Corp. v. Hulteen, 556 U.S. 701, 711-13, 129 S.Ct. 1962, 173 L.Ed.2d 898 (2009). To allow the BIA to apply Briones here, to actions during a period when Padilla-Caldera I was on the books and Briones was not the law of this circuit, would upset this principle and invite individuals to disregard on-point judicial precedent (maybe even Supreme Court precedent) because of its potential susceptibility to revision by an executive agency. See De Niz Robles, 803 F.3d at 1178-79. It would also require the people to substitute reliance on judicial declarations of what the law is with finely grained legal judgments about the degree of ambiguity in the relevant statutory provisions and the reasonableness of agency pronouncements about them. Questions that frequently provoke deep disagreement even among our most eminent jurists, let alone lay persons. See, e.g., Michigan v. EPA, — U.S. -, 135 S.Ct. 2699, 192 L.Ed.2d 674 (2015). Indeed, at the time of Mr. Gutierrez-Brizuela’s application, just as at the time of Mr. De Niz Robles’s, it was far from clear that this court would ever permit enforcement of Briones. Back then it remained a distinct possibility that this court might find either the statute unambiguously opposed to the agency’s position or the agency’s interpretation an unreasonable one. See De Niz Robles, 803 F.3d at 1179 & n.10. Are we really to expect that persons in such circumstances should have to bear the cost of ignoring directly controlling judicial precedent in favor of the speculative possibility that an executive revision might ultimately prevail? Add to that the costs associated with imposing this kind of uncertainty on an entire class of persons with significant interests at stake (the potential right to remain lawfully in this country), and the BIA’s failure to identify any benefits of any kind that might be vindicated through retroactive application of its new rule, see id. at 1179-80, and we just do not see how the agency might prevail.
Confirming our assessment on this score is the fact that the BIA itself was once— and not long ago — sensitive to the due process and equal protection concerns associated with retroactive application of its new rules. It followed a practice of defer*1148ring to adverse circuit precedent so long as that precedent remained on the books, even when the agency may have disagreed with it. See, e.g., In re Anselmo, 20 I. & N. Dec. 25, 31 (BIA 1989) (collecting cases). In fact, in Briones itself the BIA expressly refrained from deciding the applicability of its new rule to cases arising in this circuit, given that we had previously adopted a contrary view of the relevant statutes’ meaning. See Briones, 24 I. & N. Dec. at 370-71 & n.9. So it is that even the agency itself once suggested that Padilla-Caldera I would and should continue to control in this circuit until overruled by judicial authority. Given that, we are left at a loss to see how it would be consistent with due process (fair notice) principles to permit the agency now to take the opposite view. Effectively pulling out from petitioners like Mr. Gutierrez-Brizuela a rug that the agency itself set under them.
To be sure, the BIA says that was then and this is now. Now the agency replies that Briones must apply to all cases, even cases arising before its judicial adoption in circuits with contrary precedent. It must, the agency says, because “[flor Brand X to have any practical application, the Board must be able to apply Chevron step two, policy-type decisions immediately, regardless of the existence of a prior, contrary judicial interpretation. Otherwise, the Board would never be able to disagree with a prior, competing judicial decision” as Brand X says it may.
We just don’t see it. To be sure, our holding about the inefficacy of Briones to decisions predating its approval by this court in Padilla-Caldera II raises the interesting question how an agency might go about exercising its Brand X powers to overrule prior judicial decisions. If the agency tries to apply a new rule to a pending case in a circuit where there is adverse judicial precedent, and if the court replies that the judicial precedent controls until reversed, how is the agency ever to get its new authority to take effect? The question may represent yet another conundrum invited by Chevron step two and Brand X. But we have no difficulty imagining a way out of this particular one that also allows us to avoid due process and equal protection problems. Agencies could seek to enforce their new interpretations and courts could then fulfill their Chevron step two and Brand X obligation to sustain those new interpretations while also affording the agency’s new rules only prospective effect and permitting the petitioner in the case at hand to continue to receive the benefit of preexisting judicial precedent. After all, decisions like Chenery II, Stewart Capital, and Bowen expressly anticipate that there will be cases where administrative decisions should control only conduct arising after they take legal effect. See Chenery II, 332 U.S. at 203-04, 67 S.Ct. 1760; Stewart Capital, 701 F.2d at 848-50; cf. Bowen, 488 U.S. at 213-15, 109 S.Ct. 468. Neither is practice of this sort entirely foreign in other areas of the law: not infrequently courts will reach holdings that affect future litigants differently than current ones. Consider, for example, qualified immunity cases where courts sometimes find that a constitutional violation has occurred, but that the violation was not clearly established, in the process binding future litigants in a different way than current ones. See, e.g., Pearson v. Callahan, 555 U.S. 223, 129 S.Ct. 808, 172 L.Ed.2d 565 (2009); Saucier v. Katz, 533 U.S. 194, 121 S.Ct. 2151, 150 L.Ed.2d 272 (2001). Chevron step two and Brand X may mean that agencies exercising delegated legislative power can effectively overrule judicial precedents. But that does not necessarily mean their decisions must or should presumptively apply retroactively to conduct completed before they take legal effect. If anything, and as we’ve seen in De Niz Robles and again today, the opposite presumption should apply.
*1149The petition for review is granted and the case remanded to the BIA for further proceedings consistent with this opinion.
. At this point you might interject to question our analogy to legislative action. After all, an agency may enforce its new policy judgment under Chevron step two and Brand X only with judicial approval. And, quite unlike legislative action, which is subject to a presumption of prospectivity, judicial action is of course subject to a presumption of retroactivity. See De Niz Robles, 803 F.3d at 1170. But De Niz Robles already rejected an argument along these lines, see id. at 1174 n.7, and the BIA does not pursue any such argument in this case — and for good reason. The fact is, the agency rule a court ratifies in the Chevron step two/Brand X scenario clearly proceeds from the agency exercising delegated legislative authority, not the court. Consider first the fact that when announcing its rule an agency is neither seeking the "best” reading of the statutory text nor adopting a "once-and-for-always definition of what the statute means,” Garfias-Rodriguez v. Holder, 702 F.3d 504, 515-16 (9th Cir. 2012) (en banc), but seeking instead to exploit a gap in the statute to implement its own (continuously revisable) policy-influenced vision of what the law should be, De Niz Robles, 803 F.3d at 1176. Then consider the highly unusual nature of the court’s task under Chevron step two and Brand X: the court doesn’t ask whether the agency’s interpretation of the law is superior (or its prior interpretation wrong in some way), only whether the agency’s interpretation falls within the scope of the permissible. The court doesn't attempt to say what the law is (our normal function) but only whether the agency’s attempt is a reasonable one. And finally consider that, when overruling our own governing precedent, a court may not consult or speak a word about the principles of stare decisis that normally attend to (and do much to discourage) the overruling of judicial precedent. Instead, the court disregards all that and favors the agency ruling over its own precedent simply because the agency has offered a permissible interpretation. See Padilla-Caldera II, 637 F.3d at 1147. Under these circumstances, it seems to us "there can be little doubt that,” while a court must approve the agency's new rule, the rule "emanates in substance” not from the court but from the agency exercising delegated legislative authority. De Niz Robles, 803 F.3d at 1174 n.7.