It's a bit more nuanced than that. Efficacy data in mildly depressed populations is poor. In moderate depression it's marginally better. In severe depression, it clearly separates from placebo in a meaningful way.
Recent Gibbons meta-analysis in the green journal a few months ago found pretty clearly that severity really didn't modulate effect all that much. Study by Thase a little while back had similar findings.
Studies on mood relapse prevention for SSRIs is quite robust. Even when not helpful acutely, they appear very helpful for preventing recurrence.
If your criteria for "working" is that they completely cure all patients who take them (including giving somebody a job, a new girlfriend, sobriety, etc), then, yes, they're miserable.
If your criteria includes things like suicide prevention, disability, relapse prevention, etc., then our meds aren't so incomparable in effect size to plenty of interventions throughout medicine.
"Antidepressants don't work" is again a lazy, overly simplistic narrative, never mind the fact that we still have TCAs and MAOIs at our disposal which have robust effects with high side effect burdens. None of these meds make everything "all better" any more than statins prevent all heart attacks or beta blockers prevent all strokes or metformin cures diabetes or steroid inhalers cure asthma.
Nobody expects a cardiac med to give you a new heart or an inhaler to give you new lungs, so I don't know why people should expect an antidepressant to give you a new brain. Set appropriate, clinically meaningful outcomes, and there are plenty of studies that show antidepressants are useful. Prescribe fewer antidepressants, and you see the suicide rate go up. That seems like a big deal to me. The FDA gave us that little natural experiment with children after the black box warning sent pediatricians running from prescribing ADs.
Of course, if you take the easy way out and just look at meta-analyses of crappily performed pharma studies, then things don't look so good.